xref: /openbmc/linux/kernel/exit.c (revision 4a6806cfcbca2cd7bae94b2d4244dab3aaa1b333)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/kernel/exit.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 #include <linux/mm.h>
9 #include <linux/slab.h>
10 #include <linux/sched/autogroup.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/task.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/sched/cputime.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/capability.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/tty.h>
22 #include <linux/iocontext.h>
23 #include <linux/key.h>
24 #include <linux/cpu.h>
25 #include <linux/acct.h>
26 #include <linux/tsacct_kern.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/freezer.h>
30 #include <linux/binfmts.h>
31 #include <linux/nsproxy.h>
32 #include <linux/pid_namespace.h>
33 #include <linux/ptrace.h>
34 #include <linux/profile.h>
35 #include <linux/mount.h>
36 #include <linux/proc_fs.h>
37 #include <linux/kthread.h>
38 #include <linux/mempolicy.h>
39 #include <linux/taskstats_kern.h>
40 #include <linux/delayacct.h>
41 #include <linux/cgroup.h>
42 #include <linux/syscalls.h>
43 #include <linux/signal.h>
44 #include <linux/posix-timers.h>
45 #include <linux/cn_proc.h>
46 #include <linux/mutex.h>
47 #include <linux/futex.h>
48 #include <linux/pipe_fs_i.h>
49 #include <linux/audit.h> /* for audit_free() */
50 #include <linux/resource.h>
51 #include <linux/task_io_accounting_ops.h>
52 #include <linux/tracehook.h>
53 #include <linux/fs_struct.h>
54 #include <linux/init_task.h>
55 #include <linux/perf_event.h>
56 #include <trace/events/sched.h>
57 #include <linux/hw_breakpoint.h>
58 #include <linux/oom.h>
59 #include <linux/writeback.h>
60 #include <linux/shm.h>
61 #include <linux/kcov.h>
62 #include <linux/random.h>
63 #include <linux/rcuwait.h>
64 #include <linux/compat.h>
65 #include <linux/io_uring.h>
66 #include <linux/kprobes.h>
67 #include <linux/rethook.h>
68 
69 #include <linux/uaccess.h>
70 #include <asm/unistd.h>
71 #include <asm/mmu_context.h>
72 
73 static void __unhash_process(struct task_struct *p, bool group_dead)
74 {
75 	nr_threads--;
76 	detach_pid(p, PIDTYPE_PID);
77 	if (group_dead) {
78 		detach_pid(p, PIDTYPE_TGID);
79 		detach_pid(p, PIDTYPE_PGID);
80 		detach_pid(p, PIDTYPE_SID);
81 
82 		list_del_rcu(&p->tasks);
83 		list_del_init(&p->sibling);
84 		__this_cpu_dec(process_counts);
85 	}
86 	list_del_rcu(&p->thread_group);
87 	list_del_rcu(&p->thread_node);
88 }
89 
90 /*
91  * This function expects the tasklist_lock write-locked.
92  */
93 static void __exit_signal(struct task_struct *tsk)
94 {
95 	struct signal_struct *sig = tsk->signal;
96 	bool group_dead = thread_group_leader(tsk);
97 	struct sighand_struct *sighand;
98 	struct tty_struct *tty;
99 	u64 utime, stime;
100 
101 	sighand = rcu_dereference_check(tsk->sighand,
102 					lockdep_tasklist_lock_is_held());
103 	spin_lock(&sighand->siglock);
104 
105 #ifdef CONFIG_POSIX_TIMERS
106 	posix_cpu_timers_exit(tsk);
107 	if (group_dead)
108 		posix_cpu_timers_exit_group(tsk);
109 #endif
110 
111 	if (group_dead) {
112 		tty = sig->tty;
113 		sig->tty = NULL;
114 	} else {
115 		/*
116 		 * If there is any task waiting for the group exit
117 		 * then notify it:
118 		 */
119 		if (sig->notify_count > 0 && !--sig->notify_count)
120 			wake_up_process(sig->group_exec_task);
121 
122 		if (tsk == sig->curr_target)
123 			sig->curr_target = next_thread(tsk);
124 	}
125 
126 	add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
127 			      sizeof(unsigned long long));
128 
129 	/*
130 	 * Accumulate here the counters for all threads as they die. We could
131 	 * skip the group leader because it is the last user of signal_struct,
132 	 * but we want to avoid the race with thread_group_cputime() which can
133 	 * see the empty ->thread_head list.
134 	 */
135 	task_cputime(tsk, &utime, &stime);
136 	write_seqlock(&sig->stats_lock);
137 	sig->utime += utime;
138 	sig->stime += stime;
139 	sig->gtime += task_gtime(tsk);
140 	sig->min_flt += tsk->min_flt;
141 	sig->maj_flt += tsk->maj_flt;
142 	sig->nvcsw += tsk->nvcsw;
143 	sig->nivcsw += tsk->nivcsw;
144 	sig->inblock += task_io_get_inblock(tsk);
145 	sig->oublock += task_io_get_oublock(tsk);
146 	task_io_accounting_add(&sig->ioac, &tsk->ioac);
147 	sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
148 	sig->nr_threads--;
149 	__unhash_process(tsk, group_dead);
150 	write_sequnlock(&sig->stats_lock);
151 
152 	/*
153 	 * Do this under ->siglock, we can race with another thread
154 	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
155 	 */
156 	flush_sigqueue(&tsk->pending);
157 	tsk->sighand = NULL;
158 	spin_unlock(&sighand->siglock);
159 
160 	__cleanup_sighand(sighand);
161 	clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
162 	if (group_dead) {
163 		flush_sigqueue(&sig->shared_pending);
164 		tty_kref_put(tty);
165 	}
166 }
167 
168 static void delayed_put_task_struct(struct rcu_head *rhp)
169 {
170 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
171 
172 	kprobe_flush_task(tsk);
173 	rethook_flush_task(tsk);
174 	perf_event_delayed_put(tsk);
175 	trace_sched_process_free(tsk);
176 	put_task_struct(tsk);
177 }
178 
179 void put_task_struct_rcu_user(struct task_struct *task)
180 {
181 	if (refcount_dec_and_test(&task->rcu_users))
182 		call_rcu(&task->rcu, delayed_put_task_struct);
183 }
184 
185 void release_task(struct task_struct *p)
186 {
187 	struct task_struct *leader;
188 	struct pid *thread_pid;
189 	int zap_leader;
190 repeat:
191 	/* don't need to get the RCU readlock here - the process is dead and
192 	 * can't be modifying its own credentials. But shut RCU-lockdep up */
193 	rcu_read_lock();
194 	dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
195 	rcu_read_unlock();
196 
197 	cgroup_release(p);
198 
199 	write_lock_irq(&tasklist_lock);
200 	ptrace_release_task(p);
201 	thread_pid = get_pid(p->thread_pid);
202 	__exit_signal(p);
203 
204 	/*
205 	 * If we are the last non-leader member of the thread
206 	 * group, and the leader is zombie, then notify the
207 	 * group leader's parent process. (if it wants notification.)
208 	 */
209 	zap_leader = 0;
210 	leader = p->group_leader;
211 	if (leader != p && thread_group_empty(leader)
212 			&& leader->exit_state == EXIT_ZOMBIE) {
213 		/*
214 		 * If we were the last child thread and the leader has
215 		 * exited already, and the leader's parent ignores SIGCHLD,
216 		 * then we are the one who should release the leader.
217 		 */
218 		zap_leader = do_notify_parent(leader, leader->exit_signal);
219 		if (zap_leader)
220 			leader->exit_state = EXIT_DEAD;
221 	}
222 
223 	write_unlock_irq(&tasklist_lock);
224 	seccomp_filter_release(p);
225 	proc_flush_pid(thread_pid);
226 	put_pid(thread_pid);
227 	release_thread(p);
228 	put_task_struct_rcu_user(p);
229 
230 	p = leader;
231 	if (unlikely(zap_leader))
232 		goto repeat;
233 }
234 
235 int rcuwait_wake_up(struct rcuwait *w)
236 {
237 	int ret = 0;
238 	struct task_struct *task;
239 
240 	rcu_read_lock();
241 
242 	/*
243 	 * Order condition vs @task, such that everything prior to the load
244 	 * of @task is visible. This is the condition as to why the user called
245 	 * rcuwait_wake() in the first place. Pairs with set_current_state()
246 	 * barrier (A) in rcuwait_wait_event().
247 	 *
248 	 *    WAIT                WAKE
249 	 *    [S] tsk = current	  [S] cond = true
250 	 *        MB (A)	      MB (B)
251 	 *    [L] cond		  [L] tsk
252 	 */
253 	smp_mb(); /* (B) */
254 
255 	task = rcu_dereference(w->task);
256 	if (task)
257 		ret = wake_up_process(task);
258 	rcu_read_unlock();
259 
260 	return ret;
261 }
262 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
263 
264 /*
265  * Determine if a process group is "orphaned", according to the POSIX
266  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
267  * by terminal-generated stop signals.  Newly orphaned process groups are
268  * to receive a SIGHUP and a SIGCONT.
269  *
270  * "I ask you, have you ever known what it is to be an orphan?"
271  */
272 static int will_become_orphaned_pgrp(struct pid *pgrp,
273 					struct task_struct *ignored_task)
274 {
275 	struct task_struct *p;
276 
277 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
278 		if ((p == ignored_task) ||
279 		    (p->exit_state && thread_group_empty(p)) ||
280 		    is_global_init(p->real_parent))
281 			continue;
282 
283 		if (task_pgrp(p->real_parent) != pgrp &&
284 		    task_session(p->real_parent) == task_session(p))
285 			return 0;
286 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
287 
288 	return 1;
289 }
290 
291 int is_current_pgrp_orphaned(void)
292 {
293 	int retval;
294 
295 	read_lock(&tasklist_lock);
296 	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
297 	read_unlock(&tasklist_lock);
298 
299 	return retval;
300 }
301 
302 static bool has_stopped_jobs(struct pid *pgrp)
303 {
304 	struct task_struct *p;
305 
306 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
307 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
308 			return true;
309 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
310 
311 	return false;
312 }
313 
314 /*
315  * Check to see if any process groups have become orphaned as
316  * a result of our exiting, and if they have any stopped jobs,
317  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
318  */
319 static void
320 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
321 {
322 	struct pid *pgrp = task_pgrp(tsk);
323 	struct task_struct *ignored_task = tsk;
324 
325 	if (!parent)
326 		/* exit: our father is in a different pgrp than
327 		 * we are and we were the only connection outside.
328 		 */
329 		parent = tsk->real_parent;
330 	else
331 		/* reparent: our child is in a different pgrp than
332 		 * we are, and it was the only connection outside.
333 		 */
334 		ignored_task = NULL;
335 
336 	if (task_pgrp(parent) != pgrp &&
337 	    task_session(parent) == task_session(tsk) &&
338 	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
339 	    has_stopped_jobs(pgrp)) {
340 		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
341 		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
342 	}
343 }
344 
345 static void coredump_task_exit(struct task_struct *tsk)
346 {
347 	struct core_state *core_state;
348 
349 	/*
350 	 * Serialize with any possible pending coredump.
351 	 * We must hold siglock around checking core_state
352 	 * and setting PF_POSTCOREDUMP.  The core-inducing thread
353 	 * will increment ->nr_threads for each thread in the
354 	 * group without PF_POSTCOREDUMP set.
355 	 */
356 	spin_lock_irq(&tsk->sighand->siglock);
357 	tsk->flags |= PF_POSTCOREDUMP;
358 	core_state = tsk->signal->core_state;
359 	spin_unlock_irq(&tsk->sighand->siglock);
360 	if (core_state) {
361 		struct core_thread self;
362 
363 		self.task = current;
364 		if (self.task->flags & PF_SIGNALED)
365 			self.next = xchg(&core_state->dumper.next, &self);
366 		else
367 			self.task = NULL;
368 		/*
369 		 * Implies mb(), the result of xchg() must be visible
370 		 * to core_state->dumper.
371 		 */
372 		if (atomic_dec_and_test(&core_state->nr_threads))
373 			complete(&core_state->startup);
374 
375 		for (;;) {
376 			set_current_state(TASK_UNINTERRUPTIBLE);
377 			if (!self.task) /* see coredump_finish() */
378 				break;
379 			freezable_schedule();
380 		}
381 		__set_current_state(TASK_RUNNING);
382 	}
383 }
384 
385 #ifdef CONFIG_MEMCG
386 /*
387  * A task is exiting.   If it owned this mm, find a new owner for the mm.
388  */
389 void mm_update_next_owner(struct mm_struct *mm)
390 {
391 	struct task_struct *c, *g, *p = current;
392 
393 retry:
394 	/*
395 	 * If the exiting or execing task is not the owner, it's
396 	 * someone else's problem.
397 	 */
398 	if (mm->owner != p)
399 		return;
400 	/*
401 	 * The current owner is exiting/execing and there are no other
402 	 * candidates.  Do not leave the mm pointing to a possibly
403 	 * freed task structure.
404 	 */
405 	if (atomic_read(&mm->mm_users) <= 1) {
406 		WRITE_ONCE(mm->owner, NULL);
407 		return;
408 	}
409 
410 	read_lock(&tasklist_lock);
411 	/*
412 	 * Search in the children
413 	 */
414 	list_for_each_entry(c, &p->children, sibling) {
415 		if (c->mm == mm)
416 			goto assign_new_owner;
417 	}
418 
419 	/*
420 	 * Search in the siblings
421 	 */
422 	list_for_each_entry(c, &p->real_parent->children, sibling) {
423 		if (c->mm == mm)
424 			goto assign_new_owner;
425 	}
426 
427 	/*
428 	 * Search through everything else, we should not get here often.
429 	 */
430 	for_each_process(g) {
431 		if (g->flags & PF_KTHREAD)
432 			continue;
433 		for_each_thread(g, c) {
434 			if (c->mm == mm)
435 				goto assign_new_owner;
436 			if (c->mm)
437 				break;
438 		}
439 	}
440 	read_unlock(&tasklist_lock);
441 	/*
442 	 * We found no owner yet mm_users > 1: this implies that we are
443 	 * most likely racing with swapoff (try_to_unuse()) or /proc or
444 	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
445 	 */
446 	WRITE_ONCE(mm->owner, NULL);
447 	return;
448 
449 assign_new_owner:
450 	BUG_ON(c == p);
451 	get_task_struct(c);
452 	/*
453 	 * The task_lock protects c->mm from changing.
454 	 * We always want mm->owner->mm == mm
455 	 */
456 	task_lock(c);
457 	/*
458 	 * Delay read_unlock() till we have the task_lock()
459 	 * to ensure that c does not slip away underneath us
460 	 */
461 	read_unlock(&tasklist_lock);
462 	if (c->mm != mm) {
463 		task_unlock(c);
464 		put_task_struct(c);
465 		goto retry;
466 	}
467 	WRITE_ONCE(mm->owner, c);
468 	task_unlock(c);
469 	put_task_struct(c);
470 }
471 #endif /* CONFIG_MEMCG */
472 
473 /*
474  * Turn us into a lazy TLB process if we
475  * aren't already..
476  */
477 static void exit_mm(void)
478 {
479 	struct mm_struct *mm = current->mm;
480 
481 	exit_mm_release(current, mm);
482 	if (!mm)
483 		return;
484 	sync_mm_rss(mm);
485 	mmap_read_lock(mm);
486 	mmgrab(mm);
487 	BUG_ON(mm != current->active_mm);
488 	/* more a memory barrier than a real lock */
489 	task_lock(current);
490 	/*
491 	 * When a thread stops operating on an address space, the loop
492 	 * in membarrier_private_expedited() may not observe that
493 	 * tsk->mm, and the loop in membarrier_global_expedited() may
494 	 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
495 	 * rq->membarrier_state, so those would not issue an IPI.
496 	 * Membarrier requires a memory barrier after accessing
497 	 * user-space memory, before clearing tsk->mm or the
498 	 * rq->membarrier_state.
499 	 */
500 	smp_mb__after_spinlock();
501 	local_irq_disable();
502 	current->mm = NULL;
503 	membarrier_update_current_mm(NULL);
504 	enter_lazy_tlb(mm, current);
505 	local_irq_enable();
506 	task_unlock(current);
507 	mmap_read_unlock(mm);
508 	mm_update_next_owner(mm);
509 	mmput(mm);
510 	if (test_thread_flag(TIF_MEMDIE))
511 		exit_oom_victim();
512 }
513 
514 static struct task_struct *find_alive_thread(struct task_struct *p)
515 {
516 	struct task_struct *t;
517 
518 	for_each_thread(p, t) {
519 		if (!(t->flags & PF_EXITING))
520 			return t;
521 	}
522 	return NULL;
523 }
524 
525 static struct task_struct *find_child_reaper(struct task_struct *father,
526 						struct list_head *dead)
527 	__releases(&tasklist_lock)
528 	__acquires(&tasklist_lock)
529 {
530 	struct pid_namespace *pid_ns = task_active_pid_ns(father);
531 	struct task_struct *reaper = pid_ns->child_reaper;
532 	struct task_struct *p, *n;
533 
534 	if (likely(reaper != father))
535 		return reaper;
536 
537 	reaper = find_alive_thread(father);
538 	if (reaper) {
539 		pid_ns->child_reaper = reaper;
540 		return reaper;
541 	}
542 
543 	write_unlock_irq(&tasklist_lock);
544 
545 	list_for_each_entry_safe(p, n, dead, ptrace_entry) {
546 		list_del_init(&p->ptrace_entry);
547 		release_task(p);
548 	}
549 
550 	zap_pid_ns_processes(pid_ns);
551 	write_lock_irq(&tasklist_lock);
552 
553 	return father;
554 }
555 
556 /*
557  * When we die, we re-parent all our children, and try to:
558  * 1. give them to another thread in our thread group, if such a member exists
559  * 2. give it to the first ancestor process which prctl'd itself as a
560  *    child_subreaper for its children (like a service manager)
561  * 3. give it to the init process (PID 1) in our pid namespace
562  */
563 static struct task_struct *find_new_reaper(struct task_struct *father,
564 					   struct task_struct *child_reaper)
565 {
566 	struct task_struct *thread, *reaper;
567 
568 	thread = find_alive_thread(father);
569 	if (thread)
570 		return thread;
571 
572 	if (father->signal->has_child_subreaper) {
573 		unsigned int ns_level = task_pid(father)->level;
574 		/*
575 		 * Find the first ->is_child_subreaper ancestor in our pid_ns.
576 		 * We can't check reaper != child_reaper to ensure we do not
577 		 * cross the namespaces, the exiting parent could be injected
578 		 * by setns() + fork().
579 		 * We check pid->level, this is slightly more efficient than
580 		 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
581 		 */
582 		for (reaper = father->real_parent;
583 		     task_pid(reaper)->level == ns_level;
584 		     reaper = reaper->real_parent) {
585 			if (reaper == &init_task)
586 				break;
587 			if (!reaper->signal->is_child_subreaper)
588 				continue;
589 			thread = find_alive_thread(reaper);
590 			if (thread)
591 				return thread;
592 		}
593 	}
594 
595 	return child_reaper;
596 }
597 
598 /*
599 * Any that need to be release_task'd are put on the @dead list.
600  */
601 static void reparent_leader(struct task_struct *father, struct task_struct *p,
602 				struct list_head *dead)
603 {
604 	if (unlikely(p->exit_state == EXIT_DEAD))
605 		return;
606 
607 	/* We don't want people slaying init. */
608 	p->exit_signal = SIGCHLD;
609 
610 	/* If it has exited notify the new parent about this child's death. */
611 	if (!p->ptrace &&
612 	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
613 		if (do_notify_parent(p, p->exit_signal)) {
614 			p->exit_state = EXIT_DEAD;
615 			list_add(&p->ptrace_entry, dead);
616 		}
617 	}
618 
619 	kill_orphaned_pgrp(p, father);
620 }
621 
622 /*
623  * This does two things:
624  *
625  * A.  Make init inherit all the child processes
626  * B.  Check to see if any process groups have become orphaned
627  *	as a result of our exiting, and if they have any stopped
628  *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
629  */
630 static void forget_original_parent(struct task_struct *father,
631 					struct list_head *dead)
632 {
633 	struct task_struct *p, *t, *reaper;
634 
635 	if (unlikely(!list_empty(&father->ptraced)))
636 		exit_ptrace(father, dead);
637 
638 	/* Can drop and reacquire tasklist_lock */
639 	reaper = find_child_reaper(father, dead);
640 	if (list_empty(&father->children))
641 		return;
642 
643 	reaper = find_new_reaper(father, reaper);
644 	list_for_each_entry(p, &father->children, sibling) {
645 		for_each_thread(p, t) {
646 			RCU_INIT_POINTER(t->real_parent, reaper);
647 			BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
648 			if (likely(!t->ptrace))
649 				t->parent = t->real_parent;
650 			if (t->pdeath_signal)
651 				group_send_sig_info(t->pdeath_signal,
652 						    SEND_SIG_NOINFO, t,
653 						    PIDTYPE_TGID);
654 		}
655 		/*
656 		 * If this is a threaded reparent there is no need to
657 		 * notify anyone anything has happened.
658 		 */
659 		if (!same_thread_group(reaper, father))
660 			reparent_leader(father, p, dead);
661 	}
662 	list_splice_tail_init(&father->children, &reaper->children);
663 }
664 
665 /*
666  * Send signals to all our closest relatives so that they know
667  * to properly mourn us..
668  */
669 static void exit_notify(struct task_struct *tsk, int group_dead)
670 {
671 	bool autoreap;
672 	struct task_struct *p, *n;
673 	LIST_HEAD(dead);
674 
675 	write_lock_irq(&tasklist_lock);
676 	forget_original_parent(tsk, &dead);
677 
678 	if (group_dead)
679 		kill_orphaned_pgrp(tsk->group_leader, NULL);
680 
681 	tsk->exit_state = EXIT_ZOMBIE;
682 	if (unlikely(tsk->ptrace)) {
683 		int sig = thread_group_leader(tsk) &&
684 				thread_group_empty(tsk) &&
685 				!ptrace_reparented(tsk) ?
686 			tsk->exit_signal : SIGCHLD;
687 		autoreap = do_notify_parent(tsk, sig);
688 	} else if (thread_group_leader(tsk)) {
689 		autoreap = thread_group_empty(tsk) &&
690 			do_notify_parent(tsk, tsk->exit_signal);
691 	} else {
692 		autoreap = true;
693 	}
694 
695 	if (autoreap) {
696 		tsk->exit_state = EXIT_DEAD;
697 		list_add(&tsk->ptrace_entry, &dead);
698 	}
699 
700 	/* mt-exec, de_thread() is waiting for group leader */
701 	if (unlikely(tsk->signal->notify_count < 0))
702 		wake_up_process(tsk->signal->group_exec_task);
703 	write_unlock_irq(&tasklist_lock);
704 
705 	list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
706 		list_del_init(&p->ptrace_entry);
707 		release_task(p);
708 	}
709 }
710 
711 #ifdef CONFIG_DEBUG_STACK_USAGE
712 static void check_stack_usage(void)
713 {
714 	static DEFINE_SPINLOCK(low_water_lock);
715 	static int lowest_to_date = THREAD_SIZE;
716 	unsigned long free;
717 
718 	free = stack_not_used(current);
719 
720 	if (free >= lowest_to_date)
721 		return;
722 
723 	spin_lock(&low_water_lock);
724 	if (free < lowest_to_date) {
725 		pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
726 			current->comm, task_pid_nr(current), free);
727 		lowest_to_date = free;
728 	}
729 	spin_unlock(&low_water_lock);
730 }
731 #else
732 static inline void check_stack_usage(void) {}
733 #endif
734 
735 void __noreturn do_exit(long code)
736 {
737 	struct task_struct *tsk = current;
738 	int group_dead;
739 
740 	WARN_ON(blk_needs_flush_plug(tsk));
741 
742 	/*
743 	 * If do_dead is called because this processes oopsed, it's possible
744 	 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
745 	 * continuing. Amongst other possible reasons, this is to prevent
746 	 * mm_release()->clear_child_tid() from writing to a user-controlled
747 	 * kernel address.
748 	 *
749 	 * On uptodate architectures force_uaccess_begin is a noop.  On
750 	 * architectures that still have set_fs/get_fs in addition to handling
751 	 * oopses handles kernel threads that run as set_fs(KERNEL_DS) by
752 	 * default.
753 	 */
754 	force_uaccess_begin();
755 
756 	kcov_task_exit(tsk);
757 
758 	coredump_task_exit(tsk);
759 	ptrace_event(PTRACE_EVENT_EXIT, code);
760 
761 	validate_creds_for_do_exit(tsk);
762 
763 	io_uring_files_cancel();
764 	exit_signals(tsk);  /* sets PF_EXITING */
765 
766 	/* sync mm's RSS info before statistics gathering */
767 	if (tsk->mm)
768 		sync_mm_rss(tsk->mm);
769 	acct_update_integrals(tsk);
770 	group_dead = atomic_dec_and_test(&tsk->signal->live);
771 	if (group_dead) {
772 		/*
773 		 * If the last thread of global init has exited, panic
774 		 * immediately to get a useable coredump.
775 		 */
776 		if (unlikely(is_global_init(tsk)))
777 			panic("Attempted to kill init! exitcode=0x%08x\n",
778 				tsk->signal->group_exit_code ?: (int)code);
779 
780 #ifdef CONFIG_POSIX_TIMERS
781 		hrtimer_cancel(&tsk->signal->real_timer);
782 		exit_itimers(tsk->signal);
783 #endif
784 		if (tsk->mm)
785 			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
786 	}
787 	acct_collect(code, group_dead);
788 	if (group_dead)
789 		tty_audit_exit();
790 	audit_free(tsk);
791 
792 	tsk->exit_code = code;
793 	taskstats_exit(tsk, group_dead);
794 
795 	exit_mm();
796 
797 	if (group_dead)
798 		acct_process();
799 	trace_sched_process_exit(tsk);
800 
801 	exit_sem(tsk);
802 	exit_shm(tsk);
803 	exit_files(tsk);
804 	exit_fs(tsk);
805 	if (group_dead)
806 		disassociate_ctty(1);
807 	exit_task_namespaces(tsk);
808 	exit_task_work(tsk);
809 	exit_thread(tsk);
810 
811 	/*
812 	 * Flush inherited counters to the parent - before the parent
813 	 * gets woken up by child-exit notifications.
814 	 *
815 	 * because of cgroup mode, must be called before cgroup_exit()
816 	 */
817 	perf_event_exit_task(tsk);
818 
819 	sched_autogroup_exit_task(tsk);
820 	cgroup_exit(tsk);
821 
822 	/*
823 	 * FIXME: do that only when needed, using sched_exit tracepoint
824 	 */
825 	flush_ptrace_hw_breakpoint(tsk);
826 
827 	exit_tasks_rcu_start();
828 	exit_notify(tsk, group_dead);
829 	proc_exit_connector(tsk);
830 	mpol_put_task_policy(tsk);
831 #ifdef CONFIG_FUTEX
832 	if (unlikely(current->pi_state_cache))
833 		kfree(current->pi_state_cache);
834 #endif
835 	/*
836 	 * Make sure we are holding no locks:
837 	 */
838 	debug_check_no_locks_held();
839 
840 	if (tsk->io_context)
841 		exit_io_context(tsk);
842 
843 	if (tsk->splice_pipe)
844 		free_pipe_info(tsk->splice_pipe);
845 
846 	if (tsk->task_frag.page)
847 		put_page(tsk->task_frag.page);
848 
849 	validate_creds_for_do_exit(tsk);
850 
851 	check_stack_usage();
852 	preempt_disable();
853 	if (tsk->nr_dirtied)
854 		__this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
855 	exit_rcu();
856 	exit_tasks_rcu_finish();
857 
858 	lockdep_free_task(tsk);
859 	do_task_dead();
860 }
861 
862 void __noreturn make_task_dead(int signr)
863 {
864 	/*
865 	 * Take the task off the cpu after something catastrophic has
866 	 * happened.
867 	 *
868 	 * We can get here from a kernel oops, sometimes with preemption off.
869 	 * Start by checking for critical errors.
870 	 * Then fix up important state like USER_DS and preemption.
871 	 * Then do everything else.
872 	 */
873 	struct task_struct *tsk = current;
874 
875 	if (unlikely(in_interrupt()))
876 		panic("Aiee, killing interrupt handler!");
877 	if (unlikely(!tsk->pid))
878 		panic("Attempted to kill the idle task!");
879 
880 	if (unlikely(in_atomic())) {
881 		pr_info("note: %s[%d] exited with preempt_count %d\n",
882 			current->comm, task_pid_nr(current),
883 			preempt_count());
884 		preempt_count_set(PREEMPT_ENABLED);
885 	}
886 
887 	/*
888 	 * We're taking recursive faults here in make_task_dead. Safest is to just
889 	 * leave this task alone and wait for reboot.
890 	 */
891 	if (unlikely(tsk->flags & PF_EXITING)) {
892 		pr_alert("Fixing recursive fault but reboot is needed!\n");
893 		futex_exit_recursive(tsk);
894 		tsk->exit_state = EXIT_DEAD;
895 		refcount_inc(&tsk->rcu_users);
896 		do_task_dead();
897 	}
898 
899 	do_exit(signr);
900 }
901 
902 SYSCALL_DEFINE1(exit, int, error_code)
903 {
904 	do_exit((error_code&0xff)<<8);
905 }
906 
907 /*
908  * Take down every thread in the group.  This is called by fatal signals
909  * as well as by sys_exit_group (below).
910  */
911 void
912 do_group_exit(int exit_code)
913 {
914 	struct signal_struct *sig = current->signal;
915 
916 	if (sig->flags & SIGNAL_GROUP_EXIT)
917 		exit_code = sig->group_exit_code;
918 	else if (sig->group_exec_task)
919 		exit_code = 0;
920 	else if (!thread_group_empty(current)) {
921 		struct sighand_struct *const sighand = current->sighand;
922 
923 		spin_lock_irq(&sighand->siglock);
924 		if (sig->flags & SIGNAL_GROUP_EXIT)
925 			/* Another thread got here before we took the lock.  */
926 			exit_code = sig->group_exit_code;
927 		else if (sig->group_exec_task)
928 			exit_code = 0;
929 		else {
930 			sig->group_exit_code = exit_code;
931 			sig->flags = SIGNAL_GROUP_EXIT;
932 			zap_other_threads(current);
933 		}
934 		spin_unlock_irq(&sighand->siglock);
935 	}
936 
937 	do_exit(exit_code);
938 	/* NOTREACHED */
939 }
940 
941 /*
942  * this kills every thread in the thread group. Note that any externally
943  * wait4()-ing process will get the correct exit code - even if this
944  * thread is not the thread group leader.
945  */
946 SYSCALL_DEFINE1(exit_group, int, error_code)
947 {
948 	do_group_exit((error_code & 0xff) << 8);
949 	/* NOTREACHED */
950 	return 0;
951 }
952 
953 struct waitid_info {
954 	pid_t pid;
955 	uid_t uid;
956 	int status;
957 	int cause;
958 };
959 
960 struct wait_opts {
961 	enum pid_type		wo_type;
962 	int			wo_flags;
963 	struct pid		*wo_pid;
964 
965 	struct waitid_info	*wo_info;
966 	int			wo_stat;
967 	struct rusage		*wo_rusage;
968 
969 	wait_queue_entry_t		child_wait;
970 	int			notask_error;
971 };
972 
973 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
974 {
975 	return	wo->wo_type == PIDTYPE_MAX ||
976 		task_pid_type(p, wo->wo_type) == wo->wo_pid;
977 }
978 
979 static int
980 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
981 {
982 	if (!eligible_pid(wo, p))
983 		return 0;
984 
985 	/*
986 	 * Wait for all children (clone and not) if __WALL is set or
987 	 * if it is traced by us.
988 	 */
989 	if (ptrace || (wo->wo_flags & __WALL))
990 		return 1;
991 
992 	/*
993 	 * Otherwise, wait for clone children *only* if __WCLONE is set;
994 	 * otherwise, wait for non-clone children *only*.
995 	 *
996 	 * Note: a "clone" child here is one that reports to its parent
997 	 * using a signal other than SIGCHLD, or a non-leader thread which
998 	 * we can only see if it is traced by us.
999 	 */
1000 	if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1001 		return 0;
1002 
1003 	return 1;
1004 }
1005 
1006 /*
1007  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1008  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1009  * the lock and this task is uninteresting.  If we return nonzero, we have
1010  * released the lock and the system call should return.
1011  */
1012 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1013 {
1014 	int state, status;
1015 	pid_t pid = task_pid_vnr(p);
1016 	uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1017 	struct waitid_info *infop;
1018 
1019 	if (!likely(wo->wo_flags & WEXITED))
1020 		return 0;
1021 
1022 	if (unlikely(wo->wo_flags & WNOWAIT)) {
1023 		status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1024 			? p->signal->group_exit_code : p->exit_code;
1025 		get_task_struct(p);
1026 		read_unlock(&tasklist_lock);
1027 		sched_annotate_sleep();
1028 		if (wo->wo_rusage)
1029 			getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1030 		put_task_struct(p);
1031 		goto out_info;
1032 	}
1033 	/*
1034 	 * Move the task's state to DEAD/TRACE, only one thread can do this.
1035 	 */
1036 	state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1037 		EXIT_TRACE : EXIT_DEAD;
1038 	if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1039 		return 0;
1040 	/*
1041 	 * We own this thread, nobody else can reap it.
1042 	 */
1043 	read_unlock(&tasklist_lock);
1044 	sched_annotate_sleep();
1045 
1046 	/*
1047 	 * Check thread_group_leader() to exclude the traced sub-threads.
1048 	 */
1049 	if (state == EXIT_DEAD && thread_group_leader(p)) {
1050 		struct signal_struct *sig = p->signal;
1051 		struct signal_struct *psig = current->signal;
1052 		unsigned long maxrss;
1053 		u64 tgutime, tgstime;
1054 
1055 		/*
1056 		 * The resource counters for the group leader are in its
1057 		 * own task_struct.  Those for dead threads in the group
1058 		 * are in its signal_struct, as are those for the child
1059 		 * processes it has previously reaped.  All these
1060 		 * accumulate in the parent's signal_struct c* fields.
1061 		 *
1062 		 * We don't bother to take a lock here to protect these
1063 		 * p->signal fields because the whole thread group is dead
1064 		 * and nobody can change them.
1065 		 *
1066 		 * psig->stats_lock also protects us from our sub-theads
1067 		 * which can reap other children at the same time. Until
1068 		 * we change k_getrusage()-like users to rely on this lock
1069 		 * we have to take ->siglock as well.
1070 		 *
1071 		 * We use thread_group_cputime_adjusted() to get times for
1072 		 * the thread group, which consolidates times for all threads
1073 		 * in the group including the group leader.
1074 		 */
1075 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1076 		spin_lock_irq(&current->sighand->siglock);
1077 		write_seqlock(&psig->stats_lock);
1078 		psig->cutime += tgutime + sig->cutime;
1079 		psig->cstime += tgstime + sig->cstime;
1080 		psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1081 		psig->cmin_flt +=
1082 			p->min_flt + sig->min_flt + sig->cmin_flt;
1083 		psig->cmaj_flt +=
1084 			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1085 		psig->cnvcsw +=
1086 			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1087 		psig->cnivcsw +=
1088 			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1089 		psig->cinblock +=
1090 			task_io_get_inblock(p) +
1091 			sig->inblock + sig->cinblock;
1092 		psig->coublock +=
1093 			task_io_get_oublock(p) +
1094 			sig->oublock + sig->coublock;
1095 		maxrss = max(sig->maxrss, sig->cmaxrss);
1096 		if (psig->cmaxrss < maxrss)
1097 			psig->cmaxrss = maxrss;
1098 		task_io_accounting_add(&psig->ioac, &p->ioac);
1099 		task_io_accounting_add(&psig->ioac, &sig->ioac);
1100 		write_sequnlock(&psig->stats_lock);
1101 		spin_unlock_irq(&current->sighand->siglock);
1102 	}
1103 
1104 	if (wo->wo_rusage)
1105 		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1106 	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1107 		? p->signal->group_exit_code : p->exit_code;
1108 	wo->wo_stat = status;
1109 
1110 	if (state == EXIT_TRACE) {
1111 		write_lock_irq(&tasklist_lock);
1112 		/* We dropped tasklist, ptracer could die and untrace */
1113 		ptrace_unlink(p);
1114 
1115 		/* If parent wants a zombie, don't release it now */
1116 		state = EXIT_ZOMBIE;
1117 		if (do_notify_parent(p, p->exit_signal))
1118 			state = EXIT_DEAD;
1119 		p->exit_state = state;
1120 		write_unlock_irq(&tasklist_lock);
1121 	}
1122 	if (state == EXIT_DEAD)
1123 		release_task(p);
1124 
1125 out_info:
1126 	infop = wo->wo_info;
1127 	if (infop) {
1128 		if ((status & 0x7f) == 0) {
1129 			infop->cause = CLD_EXITED;
1130 			infop->status = status >> 8;
1131 		} else {
1132 			infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1133 			infop->status = status & 0x7f;
1134 		}
1135 		infop->pid = pid;
1136 		infop->uid = uid;
1137 	}
1138 
1139 	return pid;
1140 }
1141 
1142 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1143 {
1144 	if (ptrace) {
1145 		if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1146 			return &p->exit_code;
1147 	} else {
1148 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1149 			return &p->signal->group_exit_code;
1150 	}
1151 	return NULL;
1152 }
1153 
1154 /**
1155  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1156  * @wo: wait options
1157  * @ptrace: is the wait for ptrace
1158  * @p: task to wait for
1159  *
1160  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1161  *
1162  * CONTEXT:
1163  * read_lock(&tasklist_lock), which is released if return value is
1164  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1165  *
1166  * RETURNS:
1167  * 0 if wait condition didn't exist and search for other wait conditions
1168  * should continue.  Non-zero return, -errno on failure and @p's pid on
1169  * success, implies that tasklist_lock is released and wait condition
1170  * search should terminate.
1171  */
1172 static int wait_task_stopped(struct wait_opts *wo,
1173 				int ptrace, struct task_struct *p)
1174 {
1175 	struct waitid_info *infop;
1176 	int exit_code, *p_code, why;
1177 	uid_t uid = 0; /* unneeded, required by compiler */
1178 	pid_t pid;
1179 
1180 	/*
1181 	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1182 	 */
1183 	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1184 		return 0;
1185 
1186 	if (!task_stopped_code(p, ptrace))
1187 		return 0;
1188 
1189 	exit_code = 0;
1190 	spin_lock_irq(&p->sighand->siglock);
1191 
1192 	p_code = task_stopped_code(p, ptrace);
1193 	if (unlikely(!p_code))
1194 		goto unlock_sig;
1195 
1196 	exit_code = *p_code;
1197 	if (!exit_code)
1198 		goto unlock_sig;
1199 
1200 	if (!unlikely(wo->wo_flags & WNOWAIT))
1201 		*p_code = 0;
1202 
1203 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1204 unlock_sig:
1205 	spin_unlock_irq(&p->sighand->siglock);
1206 	if (!exit_code)
1207 		return 0;
1208 
1209 	/*
1210 	 * Now we are pretty sure this task is interesting.
1211 	 * Make sure it doesn't get reaped out from under us while we
1212 	 * give up the lock and then examine it below.  We don't want to
1213 	 * keep holding onto the tasklist_lock while we call getrusage and
1214 	 * possibly take page faults for user memory.
1215 	 */
1216 	get_task_struct(p);
1217 	pid = task_pid_vnr(p);
1218 	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1219 	read_unlock(&tasklist_lock);
1220 	sched_annotate_sleep();
1221 	if (wo->wo_rusage)
1222 		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1223 	put_task_struct(p);
1224 
1225 	if (likely(!(wo->wo_flags & WNOWAIT)))
1226 		wo->wo_stat = (exit_code << 8) | 0x7f;
1227 
1228 	infop = wo->wo_info;
1229 	if (infop) {
1230 		infop->cause = why;
1231 		infop->status = exit_code;
1232 		infop->pid = pid;
1233 		infop->uid = uid;
1234 	}
1235 	return pid;
1236 }
1237 
1238 /*
1239  * Handle do_wait work for one task in a live, non-stopped state.
1240  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1241  * the lock and this task is uninteresting.  If we return nonzero, we have
1242  * released the lock and the system call should return.
1243  */
1244 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1245 {
1246 	struct waitid_info *infop;
1247 	pid_t pid;
1248 	uid_t uid;
1249 
1250 	if (!unlikely(wo->wo_flags & WCONTINUED))
1251 		return 0;
1252 
1253 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1254 		return 0;
1255 
1256 	spin_lock_irq(&p->sighand->siglock);
1257 	/* Re-check with the lock held.  */
1258 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1259 		spin_unlock_irq(&p->sighand->siglock);
1260 		return 0;
1261 	}
1262 	if (!unlikely(wo->wo_flags & WNOWAIT))
1263 		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1264 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1265 	spin_unlock_irq(&p->sighand->siglock);
1266 
1267 	pid = task_pid_vnr(p);
1268 	get_task_struct(p);
1269 	read_unlock(&tasklist_lock);
1270 	sched_annotate_sleep();
1271 	if (wo->wo_rusage)
1272 		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1273 	put_task_struct(p);
1274 
1275 	infop = wo->wo_info;
1276 	if (!infop) {
1277 		wo->wo_stat = 0xffff;
1278 	} else {
1279 		infop->cause = CLD_CONTINUED;
1280 		infop->pid = pid;
1281 		infop->uid = uid;
1282 		infop->status = SIGCONT;
1283 	}
1284 	return pid;
1285 }
1286 
1287 /*
1288  * Consider @p for a wait by @parent.
1289  *
1290  * -ECHILD should be in ->notask_error before the first call.
1291  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1292  * Returns zero if the search for a child should continue;
1293  * then ->notask_error is 0 if @p is an eligible child,
1294  * or still -ECHILD.
1295  */
1296 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1297 				struct task_struct *p)
1298 {
1299 	/*
1300 	 * We can race with wait_task_zombie() from another thread.
1301 	 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1302 	 * can't confuse the checks below.
1303 	 */
1304 	int exit_state = READ_ONCE(p->exit_state);
1305 	int ret;
1306 
1307 	if (unlikely(exit_state == EXIT_DEAD))
1308 		return 0;
1309 
1310 	ret = eligible_child(wo, ptrace, p);
1311 	if (!ret)
1312 		return ret;
1313 
1314 	if (unlikely(exit_state == EXIT_TRACE)) {
1315 		/*
1316 		 * ptrace == 0 means we are the natural parent. In this case
1317 		 * we should clear notask_error, debugger will notify us.
1318 		 */
1319 		if (likely(!ptrace))
1320 			wo->notask_error = 0;
1321 		return 0;
1322 	}
1323 
1324 	if (likely(!ptrace) && unlikely(p->ptrace)) {
1325 		/*
1326 		 * If it is traced by its real parent's group, just pretend
1327 		 * the caller is ptrace_do_wait() and reap this child if it
1328 		 * is zombie.
1329 		 *
1330 		 * This also hides group stop state from real parent; otherwise
1331 		 * a single stop can be reported twice as group and ptrace stop.
1332 		 * If a ptracer wants to distinguish these two events for its
1333 		 * own children it should create a separate process which takes
1334 		 * the role of real parent.
1335 		 */
1336 		if (!ptrace_reparented(p))
1337 			ptrace = 1;
1338 	}
1339 
1340 	/* slay zombie? */
1341 	if (exit_state == EXIT_ZOMBIE) {
1342 		/* we don't reap group leaders with subthreads */
1343 		if (!delay_group_leader(p)) {
1344 			/*
1345 			 * A zombie ptracee is only visible to its ptracer.
1346 			 * Notification and reaping will be cascaded to the
1347 			 * real parent when the ptracer detaches.
1348 			 */
1349 			if (unlikely(ptrace) || likely(!p->ptrace))
1350 				return wait_task_zombie(wo, p);
1351 		}
1352 
1353 		/*
1354 		 * Allow access to stopped/continued state via zombie by
1355 		 * falling through.  Clearing of notask_error is complex.
1356 		 *
1357 		 * When !@ptrace:
1358 		 *
1359 		 * If WEXITED is set, notask_error should naturally be
1360 		 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1361 		 * so, if there are live subthreads, there are events to
1362 		 * wait for.  If all subthreads are dead, it's still safe
1363 		 * to clear - this function will be called again in finite
1364 		 * amount time once all the subthreads are released and
1365 		 * will then return without clearing.
1366 		 *
1367 		 * When @ptrace:
1368 		 *
1369 		 * Stopped state is per-task and thus can't change once the
1370 		 * target task dies.  Only continued and exited can happen.
1371 		 * Clear notask_error if WCONTINUED | WEXITED.
1372 		 */
1373 		if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1374 			wo->notask_error = 0;
1375 	} else {
1376 		/*
1377 		 * @p is alive and it's gonna stop, continue or exit, so
1378 		 * there always is something to wait for.
1379 		 */
1380 		wo->notask_error = 0;
1381 	}
1382 
1383 	/*
1384 	 * Wait for stopped.  Depending on @ptrace, different stopped state
1385 	 * is used and the two don't interact with each other.
1386 	 */
1387 	ret = wait_task_stopped(wo, ptrace, p);
1388 	if (ret)
1389 		return ret;
1390 
1391 	/*
1392 	 * Wait for continued.  There's only one continued state and the
1393 	 * ptracer can consume it which can confuse the real parent.  Don't
1394 	 * use WCONTINUED from ptracer.  You don't need or want it.
1395 	 */
1396 	return wait_task_continued(wo, p);
1397 }
1398 
1399 /*
1400  * Do the work of do_wait() for one thread in the group, @tsk.
1401  *
1402  * -ECHILD should be in ->notask_error before the first call.
1403  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1404  * Returns zero if the search for a child should continue; then
1405  * ->notask_error is 0 if there were any eligible children,
1406  * or still -ECHILD.
1407  */
1408 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1409 {
1410 	struct task_struct *p;
1411 
1412 	list_for_each_entry(p, &tsk->children, sibling) {
1413 		int ret = wait_consider_task(wo, 0, p);
1414 
1415 		if (ret)
1416 			return ret;
1417 	}
1418 
1419 	return 0;
1420 }
1421 
1422 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1423 {
1424 	struct task_struct *p;
1425 
1426 	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1427 		int ret = wait_consider_task(wo, 1, p);
1428 
1429 		if (ret)
1430 			return ret;
1431 	}
1432 
1433 	return 0;
1434 }
1435 
1436 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1437 				int sync, void *key)
1438 {
1439 	struct wait_opts *wo = container_of(wait, struct wait_opts,
1440 						child_wait);
1441 	struct task_struct *p = key;
1442 
1443 	if (!eligible_pid(wo, p))
1444 		return 0;
1445 
1446 	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1447 		return 0;
1448 
1449 	return default_wake_function(wait, mode, sync, key);
1450 }
1451 
1452 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1453 {
1454 	__wake_up_sync_key(&parent->signal->wait_chldexit,
1455 			   TASK_INTERRUPTIBLE, p);
1456 }
1457 
1458 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1459 				 struct task_struct *target)
1460 {
1461 	struct task_struct *parent =
1462 		!ptrace ? target->real_parent : target->parent;
1463 
1464 	return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1465 				     same_thread_group(current, parent));
1466 }
1467 
1468 /*
1469  * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1470  * and tracee lists to find the target task.
1471  */
1472 static int do_wait_pid(struct wait_opts *wo)
1473 {
1474 	bool ptrace;
1475 	struct task_struct *target;
1476 	int retval;
1477 
1478 	ptrace = false;
1479 	target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1480 	if (target && is_effectively_child(wo, ptrace, target)) {
1481 		retval = wait_consider_task(wo, ptrace, target);
1482 		if (retval)
1483 			return retval;
1484 	}
1485 
1486 	ptrace = true;
1487 	target = pid_task(wo->wo_pid, PIDTYPE_PID);
1488 	if (target && target->ptrace &&
1489 	    is_effectively_child(wo, ptrace, target)) {
1490 		retval = wait_consider_task(wo, ptrace, target);
1491 		if (retval)
1492 			return retval;
1493 	}
1494 
1495 	return 0;
1496 }
1497 
1498 static long do_wait(struct wait_opts *wo)
1499 {
1500 	int retval;
1501 
1502 	trace_sched_process_wait(wo->wo_pid);
1503 
1504 	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1505 	wo->child_wait.private = current;
1506 	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1507 repeat:
1508 	/*
1509 	 * If there is nothing that can match our criteria, just get out.
1510 	 * We will clear ->notask_error to zero if we see any child that
1511 	 * might later match our criteria, even if we are not able to reap
1512 	 * it yet.
1513 	 */
1514 	wo->notask_error = -ECHILD;
1515 	if ((wo->wo_type < PIDTYPE_MAX) &&
1516 	   (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1517 		goto notask;
1518 
1519 	set_current_state(TASK_INTERRUPTIBLE);
1520 	read_lock(&tasklist_lock);
1521 
1522 	if (wo->wo_type == PIDTYPE_PID) {
1523 		retval = do_wait_pid(wo);
1524 		if (retval)
1525 			goto end;
1526 	} else {
1527 		struct task_struct *tsk = current;
1528 
1529 		do {
1530 			retval = do_wait_thread(wo, tsk);
1531 			if (retval)
1532 				goto end;
1533 
1534 			retval = ptrace_do_wait(wo, tsk);
1535 			if (retval)
1536 				goto end;
1537 
1538 			if (wo->wo_flags & __WNOTHREAD)
1539 				break;
1540 		} while_each_thread(current, tsk);
1541 	}
1542 	read_unlock(&tasklist_lock);
1543 
1544 notask:
1545 	retval = wo->notask_error;
1546 	if (!retval && !(wo->wo_flags & WNOHANG)) {
1547 		retval = -ERESTARTSYS;
1548 		if (!signal_pending(current)) {
1549 			schedule();
1550 			goto repeat;
1551 		}
1552 	}
1553 end:
1554 	__set_current_state(TASK_RUNNING);
1555 	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1556 	return retval;
1557 }
1558 
1559 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1560 			  int options, struct rusage *ru)
1561 {
1562 	struct wait_opts wo;
1563 	struct pid *pid = NULL;
1564 	enum pid_type type;
1565 	long ret;
1566 	unsigned int f_flags = 0;
1567 
1568 	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1569 			__WNOTHREAD|__WCLONE|__WALL))
1570 		return -EINVAL;
1571 	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1572 		return -EINVAL;
1573 
1574 	switch (which) {
1575 	case P_ALL:
1576 		type = PIDTYPE_MAX;
1577 		break;
1578 	case P_PID:
1579 		type = PIDTYPE_PID;
1580 		if (upid <= 0)
1581 			return -EINVAL;
1582 
1583 		pid = find_get_pid(upid);
1584 		break;
1585 	case P_PGID:
1586 		type = PIDTYPE_PGID;
1587 		if (upid < 0)
1588 			return -EINVAL;
1589 
1590 		if (upid)
1591 			pid = find_get_pid(upid);
1592 		else
1593 			pid = get_task_pid(current, PIDTYPE_PGID);
1594 		break;
1595 	case P_PIDFD:
1596 		type = PIDTYPE_PID;
1597 		if (upid < 0)
1598 			return -EINVAL;
1599 
1600 		pid = pidfd_get_pid(upid, &f_flags);
1601 		if (IS_ERR(pid))
1602 			return PTR_ERR(pid);
1603 
1604 		break;
1605 	default:
1606 		return -EINVAL;
1607 	}
1608 
1609 	wo.wo_type	= type;
1610 	wo.wo_pid	= pid;
1611 	wo.wo_flags	= options;
1612 	wo.wo_info	= infop;
1613 	wo.wo_rusage	= ru;
1614 	if (f_flags & O_NONBLOCK)
1615 		wo.wo_flags |= WNOHANG;
1616 
1617 	ret = do_wait(&wo);
1618 	if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1619 		ret = -EAGAIN;
1620 
1621 	put_pid(pid);
1622 	return ret;
1623 }
1624 
1625 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1626 		infop, int, options, struct rusage __user *, ru)
1627 {
1628 	struct rusage r;
1629 	struct waitid_info info = {.status = 0};
1630 	long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1631 	int signo = 0;
1632 
1633 	if (err > 0) {
1634 		signo = SIGCHLD;
1635 		err = 0;
1636 		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1637 			return -EFAULT;
1638 	}
1639 	if (!infop)
1640 		return err;
1641 
1642 	if (!user_write_access_begin(infop, sizeof(*infop)))
1643 		return -EFAULT;
1644 
1645 	unsafe_put_user(signo, &infop->si_signo, Efault);
1646 	unsafe_put_user(0, &infop->si_errno, Efault);
1647 	unsafe_put_user(info.cause, &infop->si_code, Efault);
1648 	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1649 	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1650 	unsafe_put_user(info.status, &infop->si_status, Efault);
1651 	user_write_access_end();
1652 	return err;
1653 Efault:
1654 	user_write_access_end();
1655 	return -EFAULT;
1656 }
1657 
1658 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1659 		  struct rusage *ru)
1660 {
1661 	struct wait_opts wo;
1662 	struct pid *pid = NULL;
1663 	enum pid_type type;
1664 	long ret;
1665 
1666 	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1667 			__WNOTHREAD|__WCLONE|__WALL))
1668 		return -EINVAL;
1669 
1670 	/* -INT_MIN is not defined */
1671 	if (upid == INT_MIN)
1672 		return -ESRCH;
1673 
1674 	if (upid == -1)
1675 		type = PIDTYPE_MAX;
1676 	else if (upid < 0) {
1677 		type = PIDTYPE_PGID;
1678 		pid = find_get_pid(-upid);
1679 	} else if (upid == 0) {
1680 		type = PIDTYPE_PGID;
1681 		pid = get_task_pid(current, PIDTYPE_PGID);
1682 	} else /* upid > 0 */ {
1683 		type = PIDTYPE_PID;
1684 		pid = find_get_pid(upid);
1685 	}
1686 
1687 	wo.wo_type	= type;
1688 	wo.wo_pid	= pid;
1689 	wo.wo_flags	= options | WEXITED;
1690 	wo.wo_info	= NULL;
1691 	wo.wo_stat	= 0;
1692 	wo.wo_rusage	= ru;
1693 	ret = do_wait(&wo);
1694 	put_pid(pid);
1695 	if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1696 		ret = -EFAULT;
1697 
1698 	return ret;
1699 }
1700 
1701 int kernel_wait(pid_t pid, int *stat)
1702 {
1703 	struct wait_opts wo = {
1704 		.wo_type	= PIDTYPE_PID,
1705 		.wo_pid		= find_get_pid(pid),
1706 		.wo_flags	= WEXITED,
1707 	};
1708 	int ret;
1709 
1710 	ret = do_wait(&wo);
1711 	if (ret > 0 && wo.wo_stat)
1712 		*stat = wo.wo_stat;
1713 	put_pid(wo.wo_pid);
1714 	return ret;
1715 }
1716 
1717 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1718 		int, options, struct rusage __user *, ru)
1719 {
1720 	struct rusage r;
1721 	long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1722 
1723 	if (err > 0) {
1724 		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1725 			return -EFAULT;
1726 	}
1727 	return err;
1728 }
1729 
1730 #ifdef __ARCH_WANT_SYS_WAITPID
1731 
1732 /*
1733  * sys_waitpid() remains for compatibility. waitpid() should be
1734  * implemented by calling sys_wait4() from libc.a.
1735  */
1736 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1737 {
1738 	return kernel_wait4(pid, stat_addr, options, NULL);
1739 }
1740 
1741 #endif
1742 
1743 #ifdef CONFIG_COMPAT
1744 COMPAT_SYSCALL_DEFINE4(wait4,
1745 	compat_pid_t, pid,
1746 	compat_uint_t __user *, stat_addr,
1747 	int, options,
1748 	struct compat_rusage __user *, ru)
1749 {
1750 	struct rusage r;
1751 	long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1752 	if (err > 0) {
1753 		if (ru && put_compat_rusage(&r, ru))
1754 			return -EFAULT;
1755 	}
1756 	return err;
1757 }
1758 
1759 COMPAT_SYSCALL_DEFINE5(waitid,
1760 		int, which, compat_pid_t, pid,
1761 		struct compat_siginfo __user *, infop, int, options,
1762 		struct compat_rusage __user *, uru)
1763 {
1764 	struct rusage ru;
1765 	struct waitid_info info = {.status = 0};
1766 	long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1767 	int signo = 0;
1768 	if (err > 0) {
1769 		signo = SIGCHLD;
1770 		err = 0;
1771 		if (uru) {
1772 			/* kernel_waitid() overwrites everything in ru */
1773 			if (COMPAT_USE_64BIT_TIME)
1774 				err = copy_to_user(uru, &ru, sizeof(ru));
1775 			else
1776 				err = put_compat_rusage(&ru, uru);
1777 			if (err)
1778 				return -EFAULT;
1779 		}
1780 	}
1781 
1782 	if (!infop)
1783 		return err;
1784 
1785 	if (!user_write_access_begin(infop, sizeof(*infop)))
1786 		return -EFAULT;
1787 
1788 	unsafe_put_user(signo, &infop->si_signo, Efault);
1789 	unsafe_put_user(0, &infop->si_errno, Efault);
1790 	unsafe_put_user(info.cause, &infop->si_code, Efault);
1791 	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1792 	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1793 	unsafe_put_user(info.status, &infop->si_status, Efault);
1794 	user_write_access_end();
1795 	return err;
1796 Efault:
1797 	user_write_access_end();
1798 	return -EFAULT;
1799 }
1800 #endif
1801 
1802 /**
1803  * thread_group_exited - check that a thread group has exited
1804  * @pid: tgid of thread group to be checked.
1805  *
1806  * Test if the thread group represented by tgid has exited (all
1807  * threads are zombies, dead or completely gone).
1808  *
1809  * Return: true if the thread group has exited. false otherwise.
1810  */
1811 bool thread_group_exited(struct pid *pid)
1812 {
1813 	struct task_struct *task;
1814 	bool exited;
1815 
1816 	rcu_read_lock();
1817 	task = pid_task(pid, PIDTYPE_PID);
1818 	exited = !task ||
1819 		(READ_ONCE(task->exit_state) && thread_group_empty(task));
1820 	rcu_read_unlock();
1821 
1822 	return exited;
1823 }
1824 EXPORT_SYMBOL(thread_group_exited);
1825 
1826 __weak void abort(void)
1827 {
1828 	BUG();
1829 
1830 	/* if that doesn't kill us, halt */
1831 	panic("Oops failed to kill thread");
1832 }
1833 EXPORT_SYMBOL(abort);
1834