xref: /openbmc/linux/kernel/exit.c (revision 82003e04)
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 #include <linux/kcov.h>
57 
58 #include <asm/uaccess.h>
59 #include <asm/unistd.h>
60 #include <asm/pgtable.h>
61 #include <asm/mmu_context.h>
62 
63 static void __unhash_process(struct task_struct *p, bool group_dead)
64 {
65 	nr_threads--;
66 	detach_pid(p, PIDTYPE_PID);
67 	if (group_dead) {
68 		detach_pid(p, PIDTYPE_PGID);
69 		detach_pid(p, PIDTYPE_SID);
70 
71 		list_del_rcu(&p->tasks);
72 		list_del_init(&p->sibling);
73 		__this_cpu_dec(process_counts);
74 	}
75 	list_del_rcu(&p->thread_group);
76 	list_del_rcu(&p->thread_node);
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 
119 	/*
120 	 * Accumulate here the counters for all threads as they die. We could
121 	 * skip the group leader because it is the last user of signal_struct,
122 	 * but we want to avoid the race with thread_group_cputime() which can
123 	 * see the empty ->thread_head list.
124 	 */
125 	task_cputime(tsk, &utime, &stime);
126 	write_seqlock(&sig->stats_lock);
127 	sig->utime += utime;
128 	sig->stime += stime;
129 	sig->gtime += task_gtime(tsk);
130 	sig->min_flt += tsk->min_flt;
131 	sig->maj_flt += tsk->maj_flt;
132 	sig->nvcsw += tsk->nvcsw;
133 	sig->nivcsw += tsk->nivcsw;
134 	sig->inblock += task_io_get_inblock(tsk);
135 	sig->oublock += task_io_get_oublock(tsk);
136 	task_io_accounting_add(&sig->ioac, &tsk->ioac);
137 	sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
138 	sig->nr_threads--;
139 	__unhash_process(tsk, group_dead);
140 	write_sequnlock(&sig->stats_lock);
141 
142 	/*
143 	 * Do this under ->siglock, we can race with another thread
144 	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
145 	 */
146 	flush_sigqueue(&tsk->pending);
147 	tsk->sighand = NULL;
148 	spin_unlock(&sighand->siglock);
149 
150 	__cleanup_sighand(sighand);
151 	clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
152 	if (group_dead) {
153 		flush_sigqueue(&sig->shared_pending);
154 		tty_kref_put(tty);
155 	}
156 }
157 
158 static void delayed_put_task_struct(struct rcu_head *rhp)
159 {
160 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161 
162 	perf_event_delayed_put(tsk);
163 	trace_sched_process_free(tsk);
164 	put_task_struct(tsk);
165 }
166 
167 
168 void release_task(struct task_struct *p)
169 {
170 	struct task_struct *leader;
171 	int zap_leader;
172 repeat:
173 	/* don't need to get the RCU readlock here - the process is dead and
174 	 * can't be modifying its own credentials. But shut RCU-lockdep up */
175 	rcu_read_lock();
176 	atomic_dec(&__task_cred(p)->user->processes);
177 	rcu_read_unlock();
178 
179 	proc_flush_task(p);
180 
181 	write_lock_irq(&tasklist_lock);
182 	ptrace_release_task(p);
183 	__exit_signal(p);
184 
185 	/*
186 	 * If we are the last non-leader member of the thread
187 	 * group, and the leader is zombie, then notify the
188 	 * group leader's parent process. (if it wants notification.)
189 	 */
190 	zap_leader = 0;
191 	leader = p->group_leader;
192 	if (leader != p && thread_group_empty(leader)
193 			&& leader->exit_state == EXIT_ZOMBIE) {
194 		/*
195 		 * If we were the last child thread and the leader has
196 		 * exited already, and the leader's parent ignores SIGCHLD,
197 		 * then we are the one who should release the leader.
198 		 */
199 		zap_leader = do_notify_parent(leader, leader->exit_signal);
200 		if (zap_leader)
201 			leader->exit_state = EXIT_DEAD;
202 	}
203 
204 	write_unlock_irq(&tasklist_lock);
205 	release_thread(p);
206 	call_rcu(&p->rcu, delayed_put_task_struct);
207 
208 	p = leader;
209 	if (unlikely(zap_leader))
210 		goto repeat;
211 }
212 
213 /*
214  * Note that if this function returns a valid task_struct pointer (!NULL)
215  * task->usage must remain >0 for the duration of the RCU critical section.
216  */
217 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
218 {
219 	struct sighand_struct *sighand;
220 	struct task_struct *task;
221 
222 	/*
223 	 * We need to verify that release_task() was not called and thus
224 	 * delayed_put_task_struct() can't run and drop the last reference
225 	 * before rcu_read_unlock(). We check task->sighand != NULL,
226 	 * but we can read the already freed and reused memory.
227 	 */
228 retry:
229 	task = rcu_dereference(*ptask);
230 	if (!task)
231 		return NULL;
232 
233 	probe_kernel_address(&task->sighand, sighand);
234 
235 	/*
236 	 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
237 	 * was already freed we can not miss the preceding update of this
238 	 * pointer.
239 	 */
240 	smp_rmb();
241 	if (unlikely(task != READ_ONCE(*ptask)))
242 		goto retry;
243 
244 	/*
245 	 * We've re-checked that "task == *ptask", now we have two different
246 	 * cases:
247 	 *
248 	 * 1. This is actually the same task/task_struct. In this case
249 	 *    sighand != NULL tells us it is still alive.
250 	 *
251 	 * 2. This is another task which got the same memory for task_struct.
252 	 *    We can't know this of course, and we can not trust
253 	 *    sighand != NULL.
254 	 *
255 	 *    In this case we actually return a random value, but this is
256 	 *    correct.
257 	 *
258 	 *    If we return NULL - we can pretend that we actually noticed that
259 	 *    *ptask was updated when the previous task has exited. Or pretend
260 	 *    that probe_slab_address(&sighand) reads NULL.
261 	 *
262 	 *    If we return the new task (because sighand is not NULL for any
263 	 *    reason) - this is fine too. This (new) task can't go away before
264 	 *    another gp pass.
265 	 *
266 	 *    And note: We could even eliminate the false positive if re-read
267 	 *    task->sighand once again to avoid the falsely NULL. But this case
268 	 *    is very unlikely so we don't care.
269 	 */
270 	if (!sighand)
271 		return NULL;
272 
273 	return task;
274 }
275 
276 struct task_struct *try_get_task_struct(struct task_struct **ptask)
277 {
278 	struct task_struct *task;
279 
280 	rcu_read_lock();
281 	task = task_rcu_dereference(ptask);
282 	if (task)
283 		get_task_struct(task);
284 	rcu_read_unlock();
285 
286 	return task;
287 }
288 
289 /*
290  * Determine if a process group is "orphaned", according to the POSIX
291  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
292  * by terminal-generated stop signals.  Newly orphaned process groups are
293  * to receive a SIGHUP and a SIGCONT.
294  *
295  * "I ask you, have you ever known what it is to be an orphan?"
296  */
297 static int will_become_orphaned_pgrp(struct pid *pgrp,
298 					struct task_struct *ignored_task)
299 {
300 	struct task_struct *p;
301 
302 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
303 		if ((p == ignored_task) ||
304 		    (p->exit_state && thread_group_empty(p)) ||
305 		    is_global_init(p->real_parent))
306 			continue;
307 
308 		if (task_pgrp(p->real_parent) != pgrp &&
309 		    task_session(p->real_parent) == task_session(p))
310 			return 0;
311 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
312 
313 	return 1;
314 }
315 
316 int is_current_pgrp_orphaned(void)
317 {
318 	int retval;
319 
320 	read_lock(&tasklist_lock);
321 	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
322 	read_unlock(&tasklist_lock);
323 
324 	return retval;
325 }
326 
327 static bool has_stopped_jobs(struct pid *pgrp)
328 {
329 	struct task_struct *p;
330 
331 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
332 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
333 			return true;
334 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
335 
336 	return false;
337 }
338 
339 /*
340  * Check to see if any process groups have become orphaned as
341  * a result of our exiting, and if they have any stopped jobs,
342  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
343  */
344 static void
345 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
346 {
347 	struct pid *pgrp = task_pgrp(tsk);
348 	struct task_struct *ignored_task = tsk;
349 
350 	if (!parent)
351 		/* exit: our father is in a different pgrp than
352 		 * we are and we were the only connection outside.
353 		 */
354 		parent = tsk->real_parent;
355 	else
356 		/* reparent: our child is in a different pgrp than
357 		 * we are, and it was the only connection outside.
358 		 */
359 		ignored_task = NULL;
360 
361 	if (task_pgrp(parent) != pgrp &&
362 	    task_session(parent) == task_session(tsk) &&
363 	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
364 	    has_stopped_jobs(pgrp)) {
365 		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
366 		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
367 	}
368 }
369 
370 #ifdef CONFIG_MEMCG
371 /*
372  * A task is exiting.   If it owned this mm, find a new owner for the mm.
373  */
374 void mm_update_next_owner(struct mm_struct *mm)
375 {
376 	struct task_struct *c, *g, *p = current;
377 
378 retry:
379 	/*
380 	 * If the exiting or execing task is not the owner, it's
381 	 * someone else's problem.
382 	 */
383 	if (mm->owner != p)
384 		return;
385 	/*
386 	 * The current owner is exiting/execing and there are no other
387 	 * candidates.  Do not leave the mm pointing to a possibly
388 	 * freed task structure.
389 	 */
390 	if (atomic_read(&mm->mm_users) <= 1) {
391 		mm->owner = NULL;
392 		return;
393 	}
394 
395 	read_lock(&tasklist_lock);
396 	/*
397 	 * Search in the children
398 	 */
399 	list_for_each_entry(c, &p->children, sibling) {
400 		if (c->mm == mm)
401 			goto assign_new_owner;
402 	}
403 
404 	/*
405 	 * Search in the siblings
406 	 */
407 	list_for_each_entry(c, &p->real_parent->children, sibling) {
408 		if (c->mm == mm)
409 			goto assign_new_owner;
410 	}
411 
412 	/*
413 	 * Search through everything else, we should not get here often.
414 	 */
415 	for_each_process(g) {
416 		if (g->flags & PF_KTHREAD)
417 			continue;
418 		for_each_thread(g, c) {
419 			if (c->mm == mm)
420 				goto assign_new_owner;
421 			if (c->mm)
422 				break;
423 		}
424 	}
425 	read_unlock(&tasklist_lock);
426 	/*
427 	 * We found no owner yet mm_users > 1: this implies that we are
428 	 * most likely racing with swapoff (try_to_unuse()) or /proc or
429 	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
430 	 */
431 	mm->owner = NULL;
432 	return;
433 
434 assign_new_owner:
435 	BUG_ON(c == p);
436 	get_task_struct(c);
437 	/*
438 	 * The task_lock protects c->mm from changing.
439 	 * We always want mm->owner->mm == mm
440 	 */
441 	task_lock(c);
442 	/*
443 	 * Delay read_unlock() till we have the task_lock()
444 	 * to ensure that c does not slip away underneath us
445 	 */
446 	read_unlock(&tasklist_lock);
447 	if (c->mm != mm) {
448 		task_unlock(c);
449 		put_task_struct(c);
450 		goto retry;
451 	}
452 	mm->owner = c;
453 	task_unlock(c);
454 	put_task_struct(c);
455 }
456 #endif /* CONFIG_MEMCG */
457 
458 /*
459  * Turn us into a lazy TLB process if we
460  * aren't already..
461  */
462 static void exit_mm(struct task_struct *tsk)
463 {
464 	struct mm_struct *mm = tsk->mm;
465 	struct core_state *core_state;
466 
467 	mm_release(tsk, mm);
468 	if (!mm)
469 		return;
470 	sync_mm_rss(mm);
471 	/*
472 	 * Serialize with any possible pending coredump.
473 	 * We must hold mmap_sem around checking core_state
474 	 * and clearing tsk->mm.  The core-inducing thread
475 	 * will increment ->nr_threads for each thread in the
476 	 * group with ->mm != NULL.
477 	 */
478 	down_read(&mm->mmap_sem);
479 	core_state = mm->core_state;
480 	if (core_state) {
481 		struct core_thread self;
482 
483 		up_read(&mm->mmap_sem);
484 
485 		self.task = tsk;
486 		self.next = xchg(&core_state->dumper.next, &self);
487 		/*
488 		 * Implies mb(), the result of xchg() must be visible
489 		 * to core_state->dumper.
490 		 */
491 		if (atomic_dec_and_test(&core_state->nr_threads))
492 			complete(&core_state->startup);
493 
494 		for (;;) {
495 			set_task_state(tsk, TASK_UNINTERRUPTIBLE);
496 			if (!self.task) /* see coredump_finish() */
497 				break;
498 			freezable_schedule();
499 		}
500 		__set_task_state(tsk, TASK_RUNNING);
501 		down_read(&mm->mmap_sem);
502 	}
503 	atomic_inc(&mm->mm_count);
504 	BUG_ON(mm != tsk->active_mm);
505 	/* more a memory barrier than a real lock */
506 	task_lock(tsk);
507 	tsk->mm = NULL;
508 	up_read(&mm->mmap_sem);
509 	enter_lazy_tlb(mm, current);
510 	task_unlock(tsk);
511 	mm_update_next_owner(mm);
512 	mmput(mm);
513 	if (test_thread_flag(TIF_MEMDIE))
514 		exit_oom_victim();
515 }
516 
517 static struct task_struct *find_alive_thread(struct task_struct *p)
518 {
519 	struct task_struct *t;
520 
521 	for_each_thread(p, t) {
522 		if (!(t->flags & PF_EXITING))
523 			return t;
524 	}
525 	return NULL;
526 }
527 
528 static struct task_struct *find_child_reaper(struct task_struct *father)
529 	__releases(&tasklist_lock)
530 	__acquires(&tasklist_lock)
531 {
532 	struct pid_namespace *pid_ns = task_active_pid_ns(father);
533 	struct task_struct *reaper = pid_ns->child_reaper;
534 
535 	if (likely(reaper != father))
536 		return reaper;
537 
538 	reaper = find_alive_thread(father);
539 	if (reaper) {
540 		pid_ns->child_reaper = reaper;
541 		return reaper;
542 	}
543 
544 	write_unlock_irq(&tasklist_lock);
545 	if (unlikely(pid_ns == &init_pid_ns)) {
546 		panic("Attempted to kill init! exitcode=0x%08x\n",
547 			father->signal->group_exit_code ?: father->exit_code);
548 	}
549 	zap_pid_ns_processes(pid_ns);
550 	write_lock_irq(&tasklist_lock);
551 
552 	return father;
553 }
554 
555 /*
556  * When we die, we re-parent all our children, and try to:
557  * 1. give them to another thread in our thread group, if such a member exists
558  * 2. give it to the first ancestor process which prctl'd itself as a
559  *    child_subreaper for its children (like a service manager)
560  * 3. give it to the init process (PID 1) in our pid namespace
561  */
562 static struct task_struct *find_new_reaper(struct task_struct *father,
563 					   struct task_struct *child_reaper)
564 {
565 	struct task_struct *thread, *reaper;
566 
567 	thread = find_alive_thread(father);
568 	if (thread)
569 		return thread;
570 
571 	if (father->signal->has_child_subreaper) {
572 		/*
573 		 * Find the first ->is_child_subreaper ancestor in our pid_ns.
574 		 * We start from father to ensure we can not look into another
575 		 * namespace, this is safe because all its threads are dead.
576 		 */
577 		for (reaper = father;
578 		     !same_thread_group(reaper, child_reaper);
579 		     reaper = reaper->real_parent) {
580 			/* call_usermodehelper() descendants need this check */
581 			if (reaper == &init_task)
582 				break;
583 			if (!reaper->signal->is_child_subreaper)
584 				continue;
585 			thread = find_alive_thread(reaper);
586 			if (thread)
587 				return thread;
588 		}
589 	}
590 
591 	return child_reaper;
592 }
593 
594 /*
595 * Any that need to be release_task'd are put on the @dead list.
596  */
597 static void reparent_leader(struct task_struct *father, struct task_struct *p,
598 				struct list_head *dead)
599 {
600 	if (unlikely(p->exit_state == EXIT_DEAD))
601 		return;
602 
603 	/* We don't want people slaying init. */
604 	p->exit_signal = SIGCHLD;
605 
606 	/* If it has exited notify the new parent about this child's death. */
607 	if (!p->ptrace &&
608 	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
609 		if (do_notify_parent(p, p->exit_signal)) {
610 			p->exit_state = EXIT_DEAD;
611 			list_add(&p->ptrace_entry, dead);
612 		}
613 	}
614 
615 	kill_orphaned_pgrp(p, father);
616 }
617 
618 /*
619  * This does two things:
620  *
621  * A.  Make init inherit all the child processes
622  * B.  Check to see if any process groups have become orphaned
623  *	as a result of our exiting, and if they have any stopped
624  *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
625  */
626 static void forget_original_parent(struct task_struct *father,
627 					struct list_head *dead)
628 {
629 	struct task_struct *p, *t, *reaper;
630 
631 	if (unlikely(!list_empty(&father->ptraced)))
632 		exit_ptrace(father, dead);
633 
634 	/* Can drop and reacquire tasklist_lock */
635 	reaper = find_child_reaper(father);
636 	if (list_empty(&father->children))
637 		return;
638 
639 	reaper = find_new_reaper(father, reaper);
640 	list_for_each_entry(p, &father->children, sibling) {
641 		for_each_thread(p, t) {
642 			t->real_parent = reaper;
643 			BUG_ON((!t->ptrace) != (t->parent == father));
644 			if (likely(!t->ptrace))
645 				t->parent = t->real_parent;
646 			if (t->pdeath_signal)
647 				group_send_sig_info(t->pdeath_signal,
648 						    SEND_SIG_NOINFO, t);
649 		}
650 		/*
651 		 * If this is a threaded reparent there is no need to
652 		 * notify anyone anything has happened.
653 		 */
654 		if (!same_thread_group(reaper, father))
655 			reparent_leader(father, p, dead);
656 	}
657 	list_splice_tail_init(&father->children, &reaper->children);
658 }
659 
660 /*
661  * Send signals to all our closest relatives so that they know
662  * to properly mourn us..
663  */
664 static void exit_notify(struct task_struct *tsk, int group_dead)
665 {
666 	bool autoreap;
667 	struct task_struct *p, *n;
668 	LIST_HEAD(dead);
669 
670 	write_lock_irq(&tasklist_lock);
671 	forget_original_parent(tsk, &dead);
672 
673 	if (group_dead)
674 		kill_orphaned_pgrp(tsk->group_leader, NULL);
675 
676 	if (unlikely(tsk->ptrace)) {
677 		int sig = thread_group_leader(tsk) &&
678 				thread_group_empty(tsk) &&
679 				!ptrace_reparented(tsk) ?
680 			tsk->exit_signal : SIGCHLD;
681 		autoreap = do_notify_parent(tsk, sig);
682 	} else if (thread_group_leader(tsk)) {
683 		autoreap = thread_group_empty(tsk) &&
684 			do_notify_parent(tsk, tsk->exit_signal);
685 	} else {
686 		autoreap = true;
687 	}
688 
689 	tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
690 	if (tsk->exit_state == EXIT_DEAD)
691 		list_add(&tsk->ptrace_entry, &dead);
692 
693 	/* mt-exec, de_thread() is waiting for group leader */
694 	if (unlikely(tsk->signal->notify_count < 0))
695 		wake_up_process(tsk->signal->group_exit_task);
696 	write_unlock_irq(&tasklist_lock);
697 
698 	list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
699 		list_del_init(&p->ptrace_entry);
700 		release_task(p);
701 	}
702 }
703 
704 #ifdef CONFIG_DEBUG_STACK_USAGE
705 static void check_stack_usage(void)
706 {
707 	static DEFINE_SPINLOCK(low_water_lock);
708 	static int lowest_to_date = THREAD_SIZE;
709 	unsigned long free;
710 
711 	free = stack_not_used(current);
712 
713 	if (free >= lowest_to_date)
714 		return;
715 
716 	spin_lock(&low_water_lock);
717 	if (free < lowest_to_date) {
718 		pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
719 			current->comm, task_pid_nr(current), free);
720 		lowest_to_date = free;
721 	}
722 	spin_unlock(&low_water_lock);
723 }
724 #else
725 static inline void check_stack_usage(void) {}
726 #endif
727 
728 void __noreturn do_exit(long code)
729 {
730 	struct task_struct *tsk = current;
731 	int group_dead;
732 	TASKS_RCU(int tasks_rcu_i);
733 
734 	profile_task_exit(tsk);
735 	kcov_task_exit(tsk);
736 
737 	WARN_ON(blk_needs_flush_plug(tsk));
738 
739 	if (unlikely(in_interrupt()))
740 		panic("Aiee, killing interrupt handler!");
741 	if (unlikely(!tsk->pid))
742 		panic("Attempted to kill the idle task!");
743 
744 	/*
745 	 * If do_exit is called because this processes oopsed, it's possible
746 	 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
747 	 * continuing. Amongst other possible reasons, this is to prevent
748 	 * mm_release()->clear_child_tid() from writing to a user-controlled
749 	 * kernel address.
750 	 */
751 	set_fs(USER_DS);
752 
753 	ptrace_event(PTRACE_EVENT_EXIT, code);
754 
755 	validate_creds_for_do_exit(tsk);
756 
757 	/*
758 	 * We're taking recursive faults here in do_exit. Safest is to just
759 	 * leave this task alone and wait for reboot.
760 	 */
761 	if (unlikely(tsk->flags & PF_EXITING)) {
762 		pr_alert("Fixing recursive fault but reboot is needed!\n");
763 		/*
764 		 * We can do this unlocked here. The futex code uses
765 		 * this flag just to verify whether the pi state
766 		 * cleanup has been done or not. In the worst case it
767 		 * loops once more. We pretend that the cleanup was
768 		 * done as there is no way to return. Either the
769 		 * OWNER_DIED bit is set by now or we push the blocked
770 		 * task into the wait for ever nirwana as well.
771 		 */
772 		tsk->flags |= PF_EXITPIDONE;
773 		set_current_state(TASK_UNINTERRUPTIBLE);
774 		schedule();
775 	}
776 
777 	exit_signals(tsk);  /* sets PF_EXITING */
778 	/*
779 	 * Ensure that all new tsk->pi_lock acquisitions must observe
780 	 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
781 	 */
782 	smp_mb();
783 	/*
784 	 * Ensure that we must observe the pi_state in exit_mm() ->
785 	 * mm_release() -> exit_pi_state_list().
786 	 */
787 	raw_spin_unlock_wait(&tsk->pi_lock);
788 
789 	if (unlikely(in_atomic())) {
790 		pr_info("note: %s[%d] exited with preempt_count %d\n",
791 			current->comm, task_pid_nr(current),
792 			preempt_count());
793 		preempt_count_set(PREEMPT_ENABLED);
794 	}
795 
796 	/* sync mm's RSS info before statistics gathering */
797 	if (tsk->mm)
798 		sync_mm_rss(tsk->mm);
799 	acct_update_integrals(tsk);
800 	group_dead = atomic_dec_and_test(&tsk->signal->live);
801 	if (group_dead) {
802 		hrtimer_cancel(&tsk->signal->real_timer);
803 		exit_itimers(tsk->signal);
804 		if (tsk->mm)
805 			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
806 	}
807 	acct_collect(code, group_dead);
808 	if (group_dead)
809 		tty_audit_exit();
810 	audit_free(tsk);
811 
812 	tsk->exit_code = code;
813 	taskstats_exit(tsk, group_dead);
814 
815 	exit_mm(tsk);
816 
817 	if (group_dead)
818 		acct_process();
819 	trace_sched_process_exit(tsk);
820 
821 	exit_sem(tsk);
822 	exit_shm(tsk);
823 	exit_files(tsk);
824 	exit_fs(tsk);
825 	if (group_dead)
826 		disassociate_ctty(1);
827 	exit_task_namespaces(tsk);
828 	exit_task_work(tsk);
829 	exit_thread(tsk);
830 
831 	/*
832 	 * Flush inherited counters to the parent - before the parent
833 	 * gets woken up by child-exit notifications.
834 	 *
835 	 * because of cgroup mode, must be called before cgroup_exit()
836 	 */
837 	perf_event_exit_task(tsk);
838 
839 	cgroup_exit(tsk);
840 
841 	/*
842 	 * FIXME: do that only when needed, using sched_exit tracepoint
843 	 */
844 	flush_ptrace_hw_breakpoint(tsk);
845 
846 	TASKS_RCU(preempt_disable());
847 	TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
848 	TASKS_RCU(preempt_enable());
849 	exit_notify(tsk, group_dead);
850 	proc_exit_connector(tsk);
851 	mpol_put_task_policy(tsk);
852 #ifdef CONFIG_FUTEX
853 	if (unlikely(current->pi_state_cache))
854 		kfree(current->pi_state_cache);
855 #endif
856 	/*
857 	 * Make sure we are holding no locks:
858 	 */
859 	debug_check_no_locks_held();
860 	/*
861 	 * We can do this unlocked here. The futex code uses this flag
862 	 * just to verify whether the pi state cleanup has been done
863 	 * or not. In the worst case it loops once more.
864 	 */
865 	tsk->flags |= PF_EXITPIDONE;
866 
867 	if (tsk->io_context)
868 		exit_io_context(tsk);
869 
870 	if (tsk->splice_pipe)
871 		free_pipe_info(tsk->splice_pipe);
872 
873 	if (tsk->task_frag.page)
874 		put_page(tsk->task_frag.page);
875 
876 	validate_creds_for_do_exit(tsk);
877 
878 	check_stack_usage();
879 	preempt_disable();
880 	if (tsk->nr_dirtied)
881 		__this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
882 	exit_rcu();
883 	TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
884 
885 	do_task_dead();
886 }
887 EXPORT_SYMBOL_GPL(do_exit);
888 
889 void complete_and_exit(struct completion *comp, long code)
890 {
891 	if (comp)
892 		complete(comp);
893 
894 	do_exit(code);
895 }
896 EXPORT_SYMBOL(complete_and_exit);
897 
898 SYSCALL_DEFINE1(exit, int, error_code)
899 {
900 	do_exit((error_code&0xff)<<8);
901 }
902 
903 /*
904  * Take down every thread in the group.  This is called by fatal signals
905  * as well as by sys_exit_group (below).
906  */
907 void
908 do_group_exit(int exit_code)
909 {
910 	struct signal_struct *sig = current->signal;
911 
912 	BUG_ON(exit_code & 0x80); /* core dumps don't get here */
913 
914 	if (signal_group_exit(sig))
915 		exit_code = sig->group_exit_code;
916 	else if (!thread_group_empty(current)) {
917 		struct sighand_struct *const sighand = current->sighand;
918 
919 		spin_lock_irq(&sighand->siglock);
920 		if (signal_group_exit(sig))
921 			/* Another thread got here before we took the lock.  */
922 			exit_code = sig->group_exit_code;
923 		else {
924 			sig->group_exit_code = exit_code;
925 			sig->flags = SIGNAL_GROUP_EXIT;
926 			zap_other_threads(current);
927 		}
928 		spin_unlock_irq(&sighand->siglock);
929 	}
930 
931 	do_exit(exit_code);
932 	/* NOTREACHED */
933 }
934 
935 /*
936  * this kills every thread in the thread group. Note that any externally
937  * wait4()-ing process will get the correct exit code - even if this
938  * thread is not the thread group leader.
939  */
940 SYSCALL_DEFINE1(exit_group, int, error_code)
941 {
942 	do_group_exit((error_code & 0xff) << 8);
943 	/* NOTREACHED */
944 	return 0;
945 }
946 
947 struct wait_opts {
948 	enum pid_type		wo_type;
949 	int			wo_flags;
950 	struct pid		*wo_pid;
951 
952 	struct siginfo __user	*wo_info;
953 	int __user		*wo_stat;
954 	struct rusage __user	*wo_rusage;
955 
956 	wait_queue_t		child_wait;
957 	int			notask_error;
958 };
959 
960 static inline
961 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
962 {
963 	if (type != PIDTYPE_PID)
964 		task = task->group_leader;
965 	return task->pids[type].pid;
966 }
967 
968 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
969 {
970 	return	wo->wo_type == PIDTYPE_MAX ||
971 		task_pid_type(p, wo->wo_type) == wo->wo_pid;
972 }
973 
974 static int
975 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
976 {
977 	if (!eligible_pid(wo, p))
978 		return 0;
979 
980 	/*
981 	 * Wait for all children (clone and not) if __WALL is set or
982 	 * if it is traced by us.
983 	 */
984 	if (ptrace || (wo->wo_flags & __WALL))
985 		return 1;
986 
987 	/*
988 	 * Otherwise, wait for clone children *only* if __WCLONE is set;
989 	 * otherwise, wait for non-clone children *only*.
990 	 *
991 	 * Note: a "clone" child here is one that reports to its parent
992 	 * using a signal other than SIGCHLD, or a non-leader thread which
993 	 * we can only see if it is traced by us.
994 	 */
995 	if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
996 		return 0;
997 
998 	return 1;
999 }
1000 
1001 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1002 				pid_t pid, uid_t uid, int why, int status)
1003 {
1004 	struct siginfo __user *infop;
1005 	int retval = wo->wo_rusage
1006 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1007 
1008 	put_task_struct(p);
1009 	infop = wo->wo_info;
1010 	if (infop) {
1011 		if (!retval)
1012 			retval = put_user(SIGCHLD, &infop->si_signo);
1013 		if (!retval)
1014 			retval = put_user(0, &infop->si_errno);
1015 		if (!retval)
1016 			retval = put_user((short)why, &infop->si_code);
1017 		if (!retval)
1018 			retval = put_user(pid, &infop->si_pid);
1019 		if (!retval)
1020 			retval = put_user(uid, &infop->si_uid);
1021 		if (!retval)
1022 			retval = put_user(status, &infop->si_status);
1023 	}
1024 	if (!retval)
1025 		retval = pid;
1026 	return retval;
1027 }
1028 
1029 /*
1030  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1031  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1032  * the lock and this task is uninteresting.  If we return nonzero, we have
1033  * released the lock and the system call should return.
1034  */
1035 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1036 {
1037 	int state, retval, status;
1038 	pid_t pid = task_pid_vnr(p);
1039 	uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1040 	struct siginfo __user *infop;
1041 
1042 	if (!likely(wo->wo_flags & WEXITED))
1043 		return 0;
1044 
1045 	if (unlikely(wo->wo_flags & WNOWAIT)) {
1046 		int exit_code = p->exit_code;
1047 		int why;
1048 
1049 		get_task_struct(p);
1050 		read_unlock(&tasklist_lock);
1051 		sched_annotate_sleep();
1052 
1053 		if ((exit_code & 0x7f) == 0) {
1054 			why = CLD_EXITED;
1055 			status = exit_code >> 8;
1056 		} else {
1057 			why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1058 			status = exit_code & 0x7f;
1059 		}
1060 		return wait_noreap_copyout(wo, p, pid, uid, why, status);
1061 	}
1062 	/*
1063 	 * Move the task's state to DEAD/TRACE, only one thread can do this.
1064 	 */
1065 	state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1066 		EXIT_TRACE : EXIT_DEAD;
1067 	if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1068 		return 0;
1069 	/*
1070 	 * We own this thread, nobody else can reap it.
1071 	 */
1072 	read_unlock(&tasklist_lock);
1073 	sched_annotate_sleep();
1074 
1075 	/*
1076 	 * Check thread_group_leader() to exclude the traced sub-threads.
1077 	 */
1078 	if (state == EXIT_DEAD && thread_group_leader(p)) {
1079 		struct signal_struct *sig = p->signal;
1080 		struct signal_struct *psig = current->signal;
1081 		unsigned long maxrss;
1082 		cputime_t tgutime, tgstime;
1083 
1084 		/*
1085 		 * The resource counters for the group leader are in its
1086 		 * own task_struct.  Those for dead threads in the group
1087 		 * are in its signal_struct, as are those for the child
1088 		 * processes it has previously reaped.  All these
1089 		 * accumulate in the parent's signal_struct c* fields.
1090 		 *
1091 		 * We don't bother to take a lock here to protect these
1092 		 * p->signal fields because the whole thread group is dead
1093 		 * and nobody can change them.
1094 		 *
1095 		 * psig->stats_lock also protects us from our sub-theads
1096 		 * which can reap other children at the same time. Until
1097 		 * we change k_getrusage()-like users to rely on this lock
1098 		 * we have to take ->siglock as well.
1099 		 *
1100 		 * We use thread_group_cputime_adjusted() to get times for
1101 		 * the thread group, which consolidates times for all threads
1102 		 * in the group including the group leader.
1103 		 */
1104 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1105 		spin_lock_irq(&current->sighand->siglock);
1106 		write_seqlock(&psig->stats_lock);
1107 		psig->cutime += tgutime + sig->cutime;
1108 		psig->cstime += tgstime + sig->cstime;
1109 		psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1110 		psig->cmin_flt +=
1111 			p->min_flt + sig->min_flt + sig->cmin_flt;
1112 		psig->cmaj_flt +=
1113 			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1114 		psig->cnvcsw +=
1115 			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1116 		psig->cnivcsw +=
1117 			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1118 		psig->cinblock +=
1119 			task_io_get_inblock(p) +
1120 			sig->inblock + sig->cinblock;
1121 		psig->coublock +=
1122 			task_io_get_oublock(p) +
1123 			sig->oublock + sig->coublock;
1124 		maxrss = max(sig->maxrss, sig->cmaxrss);
1125 		if (psig->cmaxrss < maxrss)
1126 			psig->cmaxrss = maxrss;
1127 		task_io_accounting_add(&psig->ioac, &p->ioac);
1128 		task_io_accounting_add(&psig->ioac, &sig->ioac);
1129 		write_sequnlock(&psig->stats_lock);
1130 		spin_unlock_irq(&current->sighand->siglock);
1131 	}
1132 
1133 	retval = wo->wo_rusage
1134 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1135 	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1136 		? p->signal->group_exit_code : p->exit_code;
1137 	if (!retval && wo->wo_stat)
1138 		retval = put_user(status, wo->wo_stat);
1139 
1140 	infop = wo->wo_info;
1141 	if (!retval && infop)
1142 		retval = put_user(SIGCHLD, &infop->si_signo);
1143 	if (!retval && infop)
1144 		retval = put_user(0, &infop->si_errno);
1145 	if (!retval && infop) {
1146 		int why;
1147 
1148 		if ((status & 0x7f) == 0) {
1149 			why = CLD_EXITED;
1150 			status >>= 8;
1151 		} else {
1152 			why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1153 			status &= 0x7f;
1154 		}
1155 		retval = put_user((short)why, &infop->si_code);
1156 		if (!retval)
1157 			retval = put_user(status, &infop->si_status);
1158 	}
1159 	if (!retval && infop)
1160 		retval = put_user(pid, &infop->si_pid);
1161 	if (!retval && infop)
1162 		retval = put_user(uid, &infop->si_uid);
1163 	if (!retval)
1164 		retval = pid;
1165 
1166 	if (state == EXIT_TRACE) {
1167 		write_lock_irq(&tasklist_lock);
1168 		/* We dropped tasklist, ptracer could die and untrace */
1169 		ptrace_unlink(p);
1170 
1171 		/* If parent wants a zombie, don't release it now */
1172 		state = EXIT_ZOMBIE;
1173 		if (do_notify_parent(p, p->exit_signal))
1174 			state = EXIT_DEAD;
1175 		p->exit_state = state;
1176 		write_unlock_irq(&tasklist_lock);
1177 	}
1178 	if (state == EXIT_DEAD)
1179 		release_task(p);
1180 
1181 	return retval;
1182 }
1183 
1184 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1185 {
1186 	if (ptrace) {
1187 		if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1188 			return &p->exit_code;
1189 	} else {
1190 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1191 			return &p->signal->group_exit_code;
1192 	}
1193 	return NULL;
1194 }
1195 
1196 /**
1197  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1198  * @wo: wait options
1199  * @ptrace: is the wait for ptrace
1200  * @p: task to wait for
1201  *
1202  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1203  *
1204  * CONTEXT:
1205  * read_lock(&tasklist_lock), which is released if return value is
1206  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1207  *
1208  * RETURNS:
1209  * 0 if wait condition didn't exist and search for other wait conditions
1210  * should continue.  Non-zero return, -errno on failure and @p's pid on
1211  * success, implies that tasklist_lock is released and wait condition
1212  * search should terminate.
1213  */
1214 static int wait_task_stopped(struct wait_opts *wo,
1215 				int ptrace, struct task_struct *p)
1216 {
1217 	struct siginfo __user *infop;
1218 	int retval, exit_code, *p_code, why;
1219 	uid_t uid = 0; /* unneeded, required by compiler */
1220 	pid_t pid;
1221 
1222 	/*
1223 	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1224 	 */
1225 	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1226 		return 0;
1227 
1228 	if (!task_stopped_code(p, ptrace))
1229 		return 0;
1230 
1231 	exit_code = 0;
1232 	spin_lock_irq(&p->sighand->siglock);
1233 
1234 	p_code = task_stopped_code(p, ptrace);
1235 	if (unlikely(!p_code))
1236 		goto unlock_sig;
1237 
1238 	exit_code = *p_code;
1239 	if (!exit_code)
1240 		goto unlock_sig;
1241 
1242 	if (!unlikely(wo->wo_flags & WNOWAIT))
1243 		*p_code = 0;
1244 
1245 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1246 unlock_sig:
1247 	spin_unlock_irq(&p->sighand->siglock);
1248 	if (!exit_code)
1249 		return 0;
1250 
1251 	/*
1252 	 * Now we are pretty sure this task is interesting.
1253 	 * Make sure it doesn't get reaped out from under us while we
1254 	 * give up the lock and then examine it below.  We don't want to
1255 	 * keep holding onto the tasklist_lock while we call getrusage and
1256 	 * possibly take page faults for user memory.
1257 	 */
1258 	get_task_struct(p);
1259 	pid = task_pid_vnr(p);
1260 	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1261 	read_unlock(&tasklist_lock);
1262 	sched_annotate_sleep();
1263 
1264 	if (unlikely(wo->wo_flags & WNOWAIT))
1265 		return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1266 
1267 	retval = wo->wo_rusage
1268 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1269 	if (!retval && wo->wo_stat)
1270 		retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1271 
1272 	infop = wo->wo_info;
1273 	if (!retval && infop)
1274 		retval = put_user(SIGCHLD, &infop->si_signo);
1275 	if (!retval && infop)
1276 		retval = put_user(0, &infop->si_errno);
1277 	if (!retval && infop)
1278 		retval = put_user((short)why, &infop->si_code);
1279 	if (!retval && infop)
1280 		retval = put_user(exit_code, &infop->si_status);
1281 	if (!retval && infop)
1282 		retval = put_user(pid, &infop->si_pid);
1283 	if (!retval && infop)
1284 		retval = put_user(uid, &infop->si_uid);
1285 	if (!retval)
1286 		retval = pid;
1287 	put_task_struct(p);
1288 
1289 	BUG_ON(!retval);
1290 	return retval;
1291 }
1292 
1293 /*
1294  * Handle do_wait work for one task in a live, non-stopped state.
1295  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1296  * the lock and this task is uninteresting.  If we return nonzero, we have
1297  * released the lock and the system call should return.
1298  */
1299 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1300 {
1301 	int retval;
1302 	pid_t pid;
1303 	uid_t uid;
1304 
1305 	if (!unlikely(wo->wo_flags & WCONTINUED))
1306 		return 0;
1307 
1308 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1309 		return 0;
1310 
1311 	spin_lock_irq(&p->sighand->siglock);
1312 	/* Re-check with the lock held.  */
1313 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1314 		spin_unlock_irq(&p->sighand->siglock);
1315 		return 0;
1316 	}
1317 	if (!unlikely(wo->wo_flags & WNOWAIT))
1318 		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1319 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1320 	spin_unlock_irq(&p->sighand->siglock);
1321 
1322 	pid = task_pid_vnr(p);
1323 	get_task_struct(p);
1324 	read_unlock(&tasklist_lock);
1325 	sched_annotate_sleep();
1326 
1327 	if (!wo->wo_info) {
1328 		retval = wo->wo_rusage
1329 			? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1330 		put_task_struct(p);
1331 		if (!retval && wo->wo_stat)
1332 			retval = put_user(0xffff, wo->wo_stat);
1333 		if (!retval)
1334 			retval = pid;
1335 	} else {
1336 		retval = wait_noreap_copyout(wo, p, pid, uid,
1337 					     CLD_CONTINUED, SIGCONT);
1338 		BUG_ON(retval == 0);
1339 	}
1340 
1341 	return retval;
1342 }
1343 
1344 /*
1345  * Consider @p for a wait by @parent.
1346  *
1347  * -ECHILD should be in ->notask_error before the first call.
1348  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1349  * Returns zero if the search for a child should continue;
1350  * then ->notask_error is 0 if @p is an eligible child,
1351  * or another error from security_task_wait(), or still -ECHILD.
1352  */
1353 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1354 				struct task_struct *p)
1355 {
1356 	/*
1357 	 * We can race with wait_task_zombie() from another thread.
1358 	 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1359 	 * can't confuse the checks below.
1360 	 */
1361 	int exit_state = ACCESS_ONCE(p->exit_state);
1362 	int ret;
1363 
1364 	if (unlikely(exit_state == EXIT_DEAD))
1365 		return 0;
1366 
1367 	ret = eligible_child(wo, ptrace, p);
1368 	if (!ret)
1369 		return ret;
1370 
1371 	ret = security_task_wait(p);
1372 	if (unlikely(ret < 0)) {
1373 		/*
1374 		 * If we have not yet seen any eligible child,
1375 		 * then let this error code replace -ECHILD.
1376 		 * A permission error will give the user a clue
1377 		 * to look for security policy problems, rather
1378 		 * than for mysterious wait bugs.
1379 		 */
1380 		if (wo->notask_error)
1381 			wo->notask_error = ret;
1382 		return 0;
1383 	}
1384 
1385 	if (unlikely(exit_state == EXIT_TRACE)) {
1386 		/*
1387 		 * ptrace == 0 means we are the natural parent. In this case
1388 		 * we should clear notask_error, debugger will notify us.
1389 		 */
1390 		if (likely(!ptrace))
1391 			wo->notask_error = 0;
1392 		return 0;
1393 	}
1394 
1395 	if (likely(!ptrace) && unlikely(p->ptrace)) {
1396 		/*
1397 		 * If it is traced by its real parent's group, just pretend
1398 		 * the caller is ptrace_do_wait() and reap this child if it
1399 		 * is zombie.
1400 		 *
1401 		 * This also hides group stop state from real parent; otherwise
1402 		 * a single stop can be reported twice as group and ptrace stop.
1403 		 * If a ptracer wants to distinguish these two events for its
1404 		 * own children it should create a separate process which takes
1405 		 * the role of real parent.
1406 		 */
1407 		if (!ptrace_reparented(p))
1408 			ptrace = 1;
1409 	}
1410 
1411 	/* slay zombie? */
1412 	if (exit_state == EXIT_ZOMBIE) {
1413 		/* we don't reap group leaders with subthreads */
1414 		if (!delay_group_leader(p)) {
1415 			/*
1416 			 * A zombie ptracee is only visible to its ptracer.
1417 			 * Notification and reaping will be cascaded to the
1418 			 * real parent when the ptracer detaches.
1419 			 */
1420 			if (unlikely(ptrace) || likely(!p->ptrace))
1421 				return wait_task_zombie(wo, p);
1422 		}
1423 
1424 		/*
1425 		 * Allow access to stopped/continued state via zombie by
1426 		 * falling through.  Clearing of notask_error is complex.
1427 		 *
1428 		 * When !@ptrace:
1429 		 *
1430 		 * If WEXITED is set, notask_error should naturally be
1431 		 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1432 		 * so, if there are live subthreads, there are events to
1433 		 * wait for.  If all subthreads are dead, it's still safe
1434 		 * to clear - this function will be called again in finite
1435 		 * amount time once all the subthreads are released and
1436 		 * will then return without clearing.
1437 		 *
1438 		 * When @ptrace:
1439 		 *
1440 		 * Stopped state is per-task and thus can't change once the
1441 		 * target task dies.  Only continued and exited can happen.
1442 		 * Clear notask_error if WCONTINUED | WEXITED.
1443 		 */
1444 		if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1445 			wo->notask_error = 0;
1446 	} else {
1447 		/*
1448 		 * @p is alive and it's gonna stop, continue or exit, so
1449 		 * there always is something to wait for.
1450 		 */
1451 		wo->notask_error = 0;
1452 	}
1453 
1454 	/*
1455 	 * Wait for stopped.  Depending on @ptrace, different stopped state
1456 	 * is used and the two don't interact with each other.
1457 	 */
1458 	ret = wait_task_stopped(wo, ptrace, p);
1459 	if (ret)
1460 		return ret;
1461 
1462 	/*
1463 	 * Wait for continued.  There's only one continued state and the
1464 	 * ptracer can consume it which can confuse the real parent.  Don't
1465 	 * use WCONTINUED from ptracer.  You don't need or want it.
1466 	 */
1467 	return wait_task_continued(wo, p);
1468 }
1469 
1470 /*
1471  * Do the work of do_wait() for one thread in the group, @tsk.
1472  *
1473  * -ECHILD should be in ->notask_error before the first call.
1474  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1475  * Returns zero if the search for a child should continue; then
1476  * ->notask_error is 0 if there were any eligible children,
1477  * or another error from security_task_wait(), or still -ECHILD.
1478  */
1479 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1480 {
1481 	struct task_struct *p;
1482 
1483 	list_for_each_entry(p, &tsk->children, sibling) {
1484 		int ret = wait_consider_task(wo, 0, p);
1485 
1486 		if (ret)
1487 			return ret;
1488 	}
1489 
1490 	return 0;
1491 }
1492 
1493 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1494 {
1495 	struct task_struct *p;
1496 
1497 	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1498 		int ret = wait_consider_task(wo, 1, p);
1499 
1500 		if (ret)
1501 			return ret;
1502 	}
1503 
1504 	return 0;
1505 }
1506 
1507 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1508 				int sync, void *key)
1509 {
1510 	struct wait_opts *wo = container_of(wait, struct wait_opts,
1511 						child_wait);
1512 	struct task_struct *p = key;
1513 
1514 	if (!eligible_pid(wo, p))
1515 		return 0;
1516 
1517 	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1518 		return 0;
1519 
1520 	return default_wake_function(wait, mode, sync, key);
1521 }
1522 
1523 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1524 {
1525 	__wake_up_sync_key(&parent->signal->wait_chldexit,
1526 				TASK_INTERRUPTIBLE, 1, p);
1527 }
1528 
1529 static long do_wait(struct wait_opts *wo)
1530 {
1531 	struct task_struct *tsk;
1532 	int retval;
1533 
1534 	trace_sched_process_wait(wo->wo_pid);
1535 
1536 	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1537 	wo->child_wait.private = current;
1538 	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1539 repeat:
1540 	/*
1541 	 * If there is nothing that can match our criteria, just get out.
1542 	 * We will clear ->notask_error to zero if we see any child that
1543 	 * might later match our criteria, even if we are not able to reap
1544 	 * it yet.
1545 	 */
1546 	wo->notask_error = -ECHILD;
1547 	if ((wo->wo_type < PIDTYPE_MAX) &&
1548 	   (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1549 		goto notask;
1550 
1551 	set_current_state(TASK_INTERRUPTIBLE);
1552 	read_lock(&tasklist_lock);
1553 	tsk = current;
1554 	do {
1555 		retval = do_wait_thread(wo, tsk);
1556 		if (retval)
1557 			goto end;
1558 
1559 		retval = ptrace_do_wait(wo, tsk);
1560 		if (retval)
1561 			goto end;
1562 
1563 		if (wo->wo_flags & __WNOTHREAD)
1564 			break;
1565 	} while_each_thread(current, tsk);
1566 	read_unlock(&tasklist_lock);
1567 
1568 notask:
1569 	retval = wo->notask_error;
1570 	if (!retval && !(wo->wo_flags & WNOHANG)) {
1571 		retval = -ERESTARTSYS;
1572 		if (!signal_pending(current)) {
1573 			schedule();
1574 			goto repeat;
1575 		}
1576 	}
1577 end:
1578 	__set_current_state(TASK_RUNNING);
1579 	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1580 	return retval;
1581 }
1582 
1583 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1584 		infop, int, options, struct rusage __user *, ru)
1585 {
1586 	struct wait_opts wo;
1587 	struct pid *pid = NULL;
1588 	enum pid_type type;
1589 	long ret;
1590 
1591 	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1592 			__WNOTHREAD|__WCLONE|__WALL))
1593 		return -EINVAL;
1594 	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1595 		return -EINVAL;
1596 
1597 	switch (which) {
1598 	case P_ALL:
1599 		type = PIDTYPE_MAX;
1600 		break;
1601 	case P_PID:
1602 		type = PIDTYPE_PID;
1603 		if (upid <= 0)
1604 			return -EINVAL;
1605 		break;
1606 	case P_PGID:
1607 		type = PIDTYPE_PGID;
1608 		if (upid <= 0)
1609 			return -EINVAL;
1610 		break;
1611 	default:
1612 		return -EINVAL;
1613 	}
1614 
1615 	if (type < PIDTYPE_MAX)
1616 		pid = find_get_pid(upid);
1617 
1618 	wo.wo_type	= type;
1619 	wo.wo_pid	= pid;
1620 	wo.wo_flags	= options;
1621 	wo.wo_info	= infop;
1622 	wo.wo_stat	= NULL;
1623 	wo.wo_rusage	= ru;
1624 	ret = do_wait(&wo);
1625 
1626 	if (ret > 0) {
1627 		ret = 0;
1628 	} else if (infop) {
1629 		/*
1630 		 * For a WNOHANG return, clear out all the fields
1631 		 * we would set so the user can easily tell the
1632 		 * difference.
1633 		 */
1634 		if (!ret)
1635 			ret = put_user(0, &infop->si_signo);
1636 		if (!ret)
1637 			ret = put_user(0, &infop->si_errno);
1638 		if (!ret)
1639 			ret = put_user(0, &infop->si_code);
1640 		if (!ret)
1641 			ret = put_user(0, &infop->si_pid);
1642 		if (!ret)
1643 			ret = put_user(0, &infop->si_uid);
1644 		if (!ret)
1645 			ret = put_user(0, &infop->si_status);
1646 	}
1647 
1648 	put_pid(pid);
1649 	return ret;
1650 }
1651 
1652 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1653 		int, options, struct rusage __user *, ru)
1654 {
1655 	struct wait_opts wo;
1656 	struct pid *pid = NULL;
1657 	enum pid_type type;
1658 	long ret;
1659 
1660 	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1661 			__WNOTHREAD|__WCLONE|__WALL))
1662 		return -EINVAL;
1663 
1664 	if (upid == -1)
1665 		type = PIDTYPE_MAX;
1666 	else if (upid < 0) {
1667 		type = PIDTYPE_PGID;
1668 		pid = find_get_pid(-upid);
1669 	} else if (upid == 0) {
1670 		type = PIDTYPE_PGID;
1671 		pid = get_task_pid(current, PIDTYPE_PGID);
1672 	} else /* upid > 0 */ {
1673 		type = PIDTYPE_PID;
1674 		pid = find_get_pid(upid);
1675 	}
1676 
1677 	wo.wo_type	= type;
1678 	wo.wo_pid	= pid;
1679 	wo.wo_flags	= options | WEXITED;
1680 	wo.wo_info	= NULL;
1681 	wo.wo_stat	= stat_addr;
1682 	wo.wo_rusage	= ru;
1683 	ret = do_wait(&wo);
1684 	put_pid(pid);
1685 
1686 	return ret;
1687 }
1688 
1689 #ifdef __ARCH_WANT_SYS_WAITPID
1690 
1691 /*
1692  * sys_waitpid() remains for compatibility. waitpid() should be
1693  * implemented by calling sys_wait4() from libc.a.
1694  */
1695 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1696 {
1697 	return sys_wait4(pid, stat_addr, options, NULL);
1698 }
1699 
1700 #endif
1701