xref: /linux/kernel/exit.c (revision 7f71507851fc7764b36a3221839607d3a45c2025)
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/freezer.h>
29 #include <linux/binfmts.h>
30 #include <linux/nsproxy.h>
31 #include <linux/pid_namespace.h>
32 #include <linux/ptrace.h>
33 #include <linux/profile.h>
34 #include <linux/mount.h>
35 #include <linux/proc_fs.h>
36 #include <linux/kthread.h>
37 #include <linux/mempolicy.h>
38 #include <linux/taskstats_kern.h>
39 #include <linux/delayacct.h>
40 #include <linux/cgroup.h>
41 #include <linux/syscalls.h>
42 #include <linux/signal.h>
43 #include <linux/posix-timers.h>
44 #include <linux/cn_proc.h>
45 #include <linux/mutex.h>
46 #include <linux/futex.h>
47 #include <linux/pipe_fs_i.h>
48 #include <linux/audit.h> /* for audit_free() */
49 #include <linux/resource.h>
50 #include <linux/task_io_accounting_ops.h>
51 #include <linux/blkdev.h>
52 #include <linux/task_work.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/kmsan.h>
63 #include <linux/random.h>
64 #include <linux/rcuwait.h>
65 #include <linux/compat.h>
66 #include <linux/io_uring.h>
67 #include <linux/kprobes.h>
68 #include <linux/rethook.h>
69 #include <linux/sysfs.h>
70 #include <linux/user_events.h>
71 #include <linux/uaccess.h>
72 
73 #include <uapi/linux/wait.h>
74 
75 #include <asm/unistd.h>
76 #include <asm/mmu_context.h>
77 
78 #include "exit.h"
79 
80 /*
81  * The default value should be high enough to not crash a system that randomly
82  * crashes its kernel from time to time, but low enough to at least not permit
83  * overflowing 32-bit refcounts or the ldsem writer count.
84  */
85 static unsigned int oops_limit = 10000;
86 
87 #ifdef CONFIG_SYSCTL
88 static struct ctl_table kern_exit_table[] = {
89 	{
90 		.procname       = "oops_limit",
91 		.data           = &oops_limit,
92 		.maxlen         = sizeof(oops_limit),
93 		.mode           = 0644,
94 		.proc_handler   = proc_douintvec,
95 	},
96 };
97 
98 static __init int kernel_exit_sysctls_init(void)
99 {
100 	register_sysctl_init("kernel", kern_exit_table);
101 	return 0;
102 }
103 late_initcall(kernel_exit_sysctls_init);
104 #endif
105 
106 static atomic_t oops_count = ATOMIC_INIT(0);
107 
108 #ifdef CONFIG_SYSFS
109 static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
110 			       char *page)
111 {
112 	return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
113 }
114 
115 static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
116 
117 static __init int kernel_exit_sysfs_init(void)
118 {
119 	sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
120 	return 0;
121 }
122 late_initcall(kernel_exit_sysfs_init);
123 #endif
124 
125 static void __unhash_process(struct task_struct *p, bool group_dead)
126 {
127 	nr_threads--;
128 	detach_pid(p, PIDTYPE_PID);
129 	if (group_dead) {
130 		detach_pid(p, PIDTYPE_TGID);
131 		detach_pid(p, PIDTYPE_PGID);
132 		detach_pid(p, PIDTYPE_SID);
133 
134 		list_del_rcu(&p->tasks);
135 		list_del_init(&p->sibling);
136 		__this_cpu_dec(process_counts);
137 	}
138 	list_del_rcu(&p->thread_node);
139 }
140 
141 /*
142  * This function expects the tasklist_lock write-locked.
143  */
144 static void __exit_signal(struct task_struct *tsk)
145 {
146 	struct signal_struct *sig = tsk->signal;
147 	bool group_dead = thread_group_leader(tsk);
148 	struct sighand_struct *sighand;
149 	struct tty_struct *tty;
150 	u64 utime, stime;
151 
152 	sighand = rcu_dereference_check(tsk->sighand,
153 					lockdep_tasklist_lock_is_held());
154 	spin_lock(&sighand->siglock);
155 
156 #ifdef CONFIG_POSIX_TIMERS
157 	posix_cpu_timers_exit(tsk);
158 	if (group_dead)
159 		posix_cpu_timers_exit_group(tsk);
160 #endif
161 
162 	if (group_dead) {
163 		tty = sig->tty;
164 		sig->tty = NULL;
165 	} else {
166 		/*
167 		 * If there is any task waiting for the group exit
168 		 * then notify it:
169 		 */
170 		if (sig->notify_count > 0 && !--sig->notify_count)
171 			wake_up_process(sig->group_exec_task);
172 
173 		if (tsk == sig->curr_target)
174 			sig->curr_target = next_thread(tsk);
175 	}
176 
177 	add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
178 			      sizeof(unsigned long long));
179 
180 	/*
181 	 * Accumulate here the counters for all threads as they die. We could
182 	 * skip the group leader because it is the last user of signal_struct,
183 	 * but we want to avoid the race with thread_group_cputime() which can
184 	 * see the empty ->thread_head list.
185 	 */
186 	task_cputime(tsk, &utime, &stime);
187 	write_seqlock(&sig->stats_lock);
188 	sig->utime += utime;
189 	sig->stime += stime;
190 	sig->gtime += task_gtime(tsk);
191 	sig->min_flt += tsk->min_flt;
192 	sig->maj_flt += tsk->maj_flt;
193 	sig->nvcsw += tsk->nvcsw;
194 	sig->nivcsw += tsk->nivcsw;
195 	sig->inblock += task_io_get_inblock(tsk);
196 	sig->oublock += task_io_get_oublock(tsk);
197 	task_io_accounting_add(&sig->ioac, &tsk->ioac);
198 	sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
199 	sig->nr_threads--;
200 	__unhash_process(tsk, group_dead);
201 	write_sequnlock(&sig->stats_lock);
202 
203 	/*
204 	 * Do this under ->siglock, we can race with another thread
205 	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
206 	 */
207 	flush_sigqueue(&tsk->pending);
208 	tsk->sighand = NULL;
209 	spin_unlock(&sighand->siglock);
210 
211 	__cleanup_sighand(sighand);
212 	clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
213 	if (group_dead) {
214 		flush_sigqueue(&sig->shared_pending);
215 		tty_kref_put(tty);
216 	}
217 }
218 
219 static void delayed_put_task_struct(struct rcu_head *rhp)
220 {
221 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
222 
223 	kprobe_flush_task(tsk);
224 	rethook_flush_task(tsk);
225 	perf_event_delayed_put(tsk);
226 	trace_sched_process_free(tsk);
227 	put_task_struct(tsk);
228 }
229 
230 void put_task_struct_rcu_user(struct task_struct *task)
231 {
232 	if (refcount_dec_and_test(&task->rcu_users))
233 		call_rcu(&task->rcu, delayed_put_task_struct);
234 }
235 
236 void __weak release_thread(struct task_struct *dead_task)
237 {
238 }
239 
240 void release_task(struct task_struct *p)
241 {
242 	struct task_struct *leader;
243 	struct pid *thread_pid;
244 	int zap_leader;
245 repeat:
246 	/* don't need to get the RCU readlock here - the process is dead and
247 	 * can't be modifying its own credentials. But shut RCU-lockdep up */
248 	rcu_read_lock();
249 	dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
250 	rcu_read_unlock();
251 
252 	cgroup_release(p);
253 
254 	write_lock_irq(&tasklist_lock);
255 	ptrace_release_task(p);
256 	thread_pid = get_pid(p->thread_pid);
257 	__exit_signal(p);
258 
259 	/*
260 	 * If we are the last non-leader member of the thread
261 	 * group, and the leader is zombie, then notify the
262 	 * group leader's parent process. (if it wants notification.)
263 	 */
264 	zap_leader = 0;
265 	leader = p->group_leader;
266 	if (leader != p && thread_group_empty(leader)
267 			&& leader->exit_state == EXIT_ZOMBIE) {
268 		/*
269 		 * If we were the last child thread and the leader has
270 		 * exited already, and the leader's parent ignores SIGCHLD,
271 		 * then we are the one who should release the leader.
272 		 */
273 		zap_leader = do_notify_parent(leader, leader->exit_signal);
274 		if (zap_leader)
275 			leader->exit_state = EXIT_DEAD;
276 	}
277 
278 	write_unlock_irq(&tasklist_lock);
279 	proc_flush_pid(thread_pid);
280 	put_pid(thread_pid);
281 	release_thread(p);
282 	put_task_struct_rcu_user(p);
283 
284 	p = leader;
285 	if (unlikely(zap_leader))
286 		goto repeat;
287 }
288 
289 int rcuwait_wake_up(struct rcuwait *w)
290 {
291 	int ret = 0;
292 	struct task_struct *task;
293 
294 	rcu_read_lock();
295 
296 	/*
297 	 * Order condition vs @task, such that everything prior to the load
298 	 * of @task is visible. This is the condition as to why the user called
299 	 * rcuwait_wake() in the first place. Pairs with set_current_state()
300 	 * barrier (A) in rcuwait_wait_event().
301 	 *
302 	 *    WAIT                WAKE
303 	 *    [S] tsk = current	  [S] cond = true
304 	 *        MB (A)	      MB (B)
305 	 *    [L] cond		  [L] tsk
306 	 */
307 	smp_mb(); /* (B) */
308 
309 	task = rcu_dereference(w->task);
310 	if (task)
311 		ret = wake_up_process(task);
312 	rcu_read_unlock();
313 
314 	return ret;
315 }
316 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
317 
318 /*
319  * Determine if a process group is "orphaned", according to the POSIX
320  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
321  * by terminal-generated stop signals.  Newly orphaned process groups are
322  * to receive a SIGHUP and a SIGCONT.
323  *
324  * "I ask you, have you ever known what it is to be an orphan?"
325  */
326 static int will_become_orphaned_pgrp(struct pid *pgrp,
327 					struct task_struct *ignored_task)
328 {
329 	struct task_struct *p;
330 
331 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
332 		if ((p == ignored_task) ||
333 		    (p->exit_state && thread_group_empty(p)) ||
334 		    is_global_init(p->real_parent))
335 			continue;
336 
337 		if (task_pgrp(p->real_parent) != pgrp &&
338 		    task_session(p->real_parent) == task_session(p))
339 			return 0;
340 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
341 
342 	return 1;
343 }
344 
345 int is_current_pgrp_orphaned(void)
346 {
347 	int retval;
348 
349 	read_lock(&tasklist_lock);
350 	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
351 	read_unlock(&tasklist_lock);
352 
353 	return retval;
354 }
355 
356 static bool has_stopped_jobs(struct pid *pgrp)
357 {
358 	struct task_struct *p;
359 
360 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
361 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
362 			return true;
363 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
364 
365 	return false;
366 }
367 
368 /*
369  * Check to see if any process groups have become orphaned as
370  * a result of our exiting, and if they have any stopped jobs,
371  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
372  */
373 static void
374 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
375 {
376 	struct pid *pgrp = task_pgrp(tsk);
377 	struct task_struct *ignored_task = tsk;
378 
379 	if (!parent)
380 		/* exit: our father is in a different pgrp than
381 		 * we are and we were the only connection outside.
382 		 */
383 		parent = tsk->real_parent;
384 	else
385 		/* reparent: our child is in a different pgrp than
386 		 * we are, and it was the only connection outside.
387 		 */
388 		ignored_task = NULL;
389 
390 	if (task_pgrp(parent) != pgrp &&
391 	    task_session(parent) == task_session(tsk) &&
392 	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
393 	    has_stopped_jobs(pgrp)) {
394 		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
395 		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
396 	}
397 }
398 
399 static void coredump_task_exit(struct task_struct *tsk)
400 {
401 	struct core_state *core_state;
402 
403 	/*
404 	 * Serialize with any possible pending coredump.
405 	 * We must hold siglock around checking core_state
406 	 * and setting PF_POSTCOREDUMP.  The core-inducing thread
407 	 * will increment ->nr_threads for each thread in the
408 	 * group without PF_POSTCOREDUMP set.
409 	 */
410 	spin_lock_irq(&tsk->sighand->siglock);
411 	tsk->flags |= PF_POSTCOREDUMP;
412 	core_state = tsk->signal->core_state;
413 	spin_unlock_irq(&tsk->sighand->siglock);
414 	if (core_state) {
415 		struct core_thread self;
416 
417 		self.task = current;
418 		if (self.task->flags & PF_SIGNALED)
419 			self.next = xchg(&core_state->dumper.next, &self);
420 		else
421 			self.task = NULL;
422 		/*
423 		 * Implies mb(), the result of xchg() must be visible
424 		 * to core_state->dumper.
425 		 */
426 		if (atomic_dec_and_test(&core_state->nr_threads))
427 			complete(&core_state->startup);
428 
429 		for (;;) {
430 			set_current_state(TASK_IDLE|TASK_FREEZABLE);
431 			if (!self.task) /* see coredump_finish() */
432 				break;
433 			schedule();
434 		}
435 		__set_current_state(TASK_RUNNING);
436 	}
437 }
438 
439 #ifdef CONFIG_MEMCG
440 /* drops tasklist_lock if succeeds */
441 static bool __try_to_set_owner(struct task_struct *tsk, struct mm_struct *mm)
442 {
443 	bool ret = false;
444 
445 	task_lock(tsk);
446 	if (likely(tsk->mm == mm)) {
447 		/* tsk can't pass exit_mm/exec_mmap and exit */
448 		read_unlock(&tasklist_lock);
449 		WRITE_ONCE(mm->owner, tsk);
450 		lru_gen_migrate_mm(mm);
451 		ret = true;
452 	}
453 	task_unlock(tsk);
454 	return ret;
455 }
456 
457 static bool try_to_set_owner(struct task_struct *g, struct mm_struct *mm)
458 {
459 	struct task_struct *t;
460 
461 	for_each_thread(g, t) {
462 		struct mm_struct *t_mm = READ_ONCE(t->mm);
463 		if (t_mm == mm) {
464 			if (__try_to_set_owner(t, mm))
465 				return true;
466 		} else if (t_mm)
467 			break;
468 	}
469 
470 	return false;
471 }
472 
473 /*
474  * A task is exiting.   If it owned this mm, find a new owner for the mm.
475  */
476 void mm_update_next_owner(struct mm_struct *mm)
477 {
478 	struct task_struct *g, *p = current;
479 
480 	/*
481 	 * If the exiting or execing task is not the owner, it's
482 	 * someone else's problem.
483 	 */
484 	if (mm->owner != p)
485 		return;
486 	/*
487 	 * The current owner is exiting/execing and there are no other
488 	 * candidates.  Do not leave the mm pointing to a possibly
489 	 * freed task structure.
490 	 */
491 	if (atomic_read(&mm->mm_users) <= 1) {
492 		WRITE_ONCE(mm->owner, NULL);
493 		return;
494 	}
495 
496 	read_lock(&tasklist_lock);
497 	/*
498 	 * Search in the children
499 	 */
500 	list_for_each_entry(g, &p->children, sibling) {
501 		if (try_to_set_owner(g, mm))
502 			goto ret;
503 	}
504 	/*
505 	 * Search in the siblings
506 	 */
507 	list_for_each_entry(g, &p->real_parent->children, sibling) {
508 		if (try_to_set_owner(g, mm))
509 			goto ret;
510 	}
511 	/*
512 	 * Search through everything else, we should not get here often.
513 	 */
514 	for_each_process(g) {
515 		if (atomic_read(&mm->mm_users) <= 1)
516 			break;
517 		if (g->flags & PF_KTHREAD)
518 			continue;
519 		if (try_to_set_owner(g, mm))
520 			goto ret;
521 	}
522 	read_unlock(&tasklist_lock);
523 	/*
524 	 * We found no owner yet mm_users > 1: this implies that we are
525 	 * most likely racing with swapoff (try_to_unuse()) or /proc or
526 	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
527 	 */
528 	WRITE_ONCE(mm->owner, NULL);
529  ret:
530 	return;
531 
532 }
533 #endif /* CONFIG_MEMCG */
534 
535 /*
536  * Turn us into a lazy TLB process if we
537  * aren't already..
538  */
539 static void exit_mm(void)
540 {
541 	struct mm_struct *mm = current->mm;
542 
543 	exit_mm_release(current, mm);
544 	if (!mm)
545 		return;
546 	mmap_read_lock(mm);
547 	mmgrab_lazy_tlb(mm);
548 	BUG_ON(mm != current->active_mm);
549 	/* more a memory barrier than a real lock */
550 	task_lock(current);
551 	/*
552 	 * When a thread stops operating on an address space, the loop
553 	 * in membarrier_private_expedited() may not observe that
554 	 * tsk->mm, and the loop in membarrier_global_expedited() may
555 	 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
556 	 * rq->membarrier_state, so those would not issue an IPI.
557 	 * Membarrier requires a memory barrier after accessing
558 	 * user-space memory, before clearing tsk->mm or the
559 	 * rq->membarrier_state.
560 	 */
561 	smp_mb__after_spinlock();
562 	local_irq_disable();
563 	current->mm = NULL;
564 	membarrier_update_current_mm(NULL);
565 	enter_lazy_tlb(mm, current);
566 	local_irq_enable();
567 	task_unlock(current);
568 	mmap_read_unlock(mm);
569 	mm_update_next_owner(mm);
570 	mmput(mm);
571 	if (test_thread_flag(TIF_MEMDIE))
572 		exit_oom_victim();
573 }
574 
575 static struct task_struct *find_alive_thread(struct task_struct *p)
576 {
577 	struct task_struct *t;
578 
579 	for_each_thread(p, t) {
580 		if (!(t->flags & PF_EXITING))
581 			return t;
582 	}
583 	return NULL;
584 }
585 
586 static struct task_struct *find_child_reaper(struct task_struct *father,
587 						struct list_head *dead)
588 	__releases(&tasklist_lock)
589 	__acquires(&tasklist_lock)
590 {
591 	struct pid_namespace *pid_ns = task_active_pid_ns(father);
592 	struct task_struct *reaper = pid_ns->child_reaper;
593 	struct task_struct *p, *n;
594 
595 	if (likely(reaper != father))
596 		return reaper;
597 
598 	reaper = find_alive_thread(father);
599 	if (reaper) {
600 		pid_ns->child_reaper = reaper;
601 		return reaper;
602 	}
603 
604 	write_unlock_irq(&tasklist_lock);
605 
606 	list_for_each_entry_safe(p, n, dead, ptrace_entry) {
607 		list_del_init(&p->ptrace_entry);
608 		release_task(p);
609 	}
610 
611 	zap_pid_ns_processes(pid_ns);
612 	write_lock_irq(&tasklist_lock);
613 
614 	return father;
615 }
616 
617 /*
618  * When we die, we re-parent all our children, and try to:
619  * 1. give them to another thread in our thread group, if such a member exists
620  * 2. give it to the first ancestor process which prctl'd itself as a
621  *    child_subreaper for its children (like a service manager)
622  * 3. give it to the init process (PID 1) in our pid namespace
623  */
624 static struct task_struct *find_new_reaper(struct task_struct *father,
625 					   struct task_struct *child_reaper)
626 {
627 	struct task_struct *thread, *reaper;
628 
629 	thread = find_alive_thread(father);
630 	if (thread)
631 		return thread;
632 
633 	if (father->signal->has_child_subreaper) {
634 		unsigned int ns_level = task_pid(father)->level;
635 		/*
636 		 * Find the first ->is_child_subreaper ancestor in our pid_ns.
637 		 * We can't check reaper != child_reaper to ensure we do not
638 		 * cross the namespaces, the exiting parent could be injected
639 		 * by setns() + fork().
640 		 * We check pid->level, this is slightly more efficient than
641 		 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
642 		 */
643 		for (reaper = father->real_parent;
644 		     task_pid(reaper)->level == ns_level;
645 		     reaper = reaper->real_parent) {
646 			if (reaper == &init_task)
647 				break;
648 			if (!reaper->signal->is_child_subreaper)
649 				continue;
650 			thread = find_alive_thread(reaper);
651 			if (thread)
652 				return thread;
653 		}
654 	}
655 
656 	return child_reaper;
657 }
658 
659 /*
660 * Any that need to be release_task'd are put on the @dead list.
661  */
662 static void reparent_leader(struct task_struct *father, struct task_struct *p,
663 				struct list_head *dead)
664 {
665 	if (unlikely(p->exit_state == EXIT_DEAD))
666 		return;
667 
668 	/* We don't want people slaying init. */
669 	p->exit_signal = SIGCHLD;
670 
671 	/* If it has exited notify the new parent about this child's death. */
672 	if (!p->ptrace &&
673 	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
674 		if (do_notify_parent(p, p->exit_signal)) {
675 			p->exit_state = EXIT_DEAD;
676 			list_add(&p->ptrace_entry, dead);
677 		}
678 	}
679 
680 	kill_orphaned_pgrp(p, father);
681 }
682 
683 /*
684  * This does two things:
685  *
686  * A.  Make init inherit all the child processes
687  * B.  Check to see if any process groups have become orphaned
688  *	as a result of our exiting, and if they have any stopped
689  *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
690  */
691 static void forget_original_parent(struct task_struct *father,
692 					struct list_head *dead)
693 {
694 	struct task_struct *p, *t, *reaper;
695 
696 	if (unlikely(!list_empty(&father->ptraced)))
697 		exit_ptrace(father, dead);
698 
699 	/* Can drop and reacquire tasklist_lock */
700 	reaper = find_child_reaper(father, dead);
701 	if (list_empty(&father->children))
702 		return;
703 
704 	reaper = find_new_reaper(father, reaper);
705 	list_for_each_entry(p, &father->children, sibling) {
706 		for_each_thread(p, t) {
707 			RCU_INIT_POINTER(t->real_parent, reaper);
708 			BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
709 			if (likely(!t->ptrace))
710 				t->parent = t->real_parent;
711 			if (t->pdeath_signal)
712 				group_send_sig_info(t->pdeath_signal,
713 						    SEND_SIG_NOINFO, t,
714 						    PIDTYPE_TGID);
715 		}
716 		/*
717 		 * If this is a threaded reparent there is no need to
718 		 * notify anyone anything has happened.
719 		 */
720 		if (!same_thread_group(reaper, father))
721 			reparent_leader(father, p, dead);
722 	}
723 	list_splice_tail_init(&father->children, &reaper->children);
724 }
725 
726 /*
727  * Send signals to all our closest relatives so that they know
728  * to properly mourn us..
729  */
730 static void exit_notify(struct task_struct *tsk, int group_dead)
731 {
732 	bool autoreap;
733 	struct task_struct *p, *n;
734 	LIST_HEAD(dead);
735 
736 	write_lock_irq(&tasklist_lock);
737 	forget_original_parent(tsk, &dead);
738 
739 	if (group_dead)
740 		kill_orphaned_pgrp(tsk->group_leader, NULL);
741 
742 	tsk->exit_state = EXIT_ZOMBIE;
743 	/*
744 	 * sub-thread or delay_group_leader(), wake up the
745 	 * PIDFD_THREAD waiters.
746 	 */
747 	if (!thread_group_empty(tsk))
748 		do_notify_pidfd(tsk);
749 
750 	if (unlikely(tsk->ptrace)) {
751 		int sig = thread_group_leader(tsk) &&
752 				thread_group_empty(tsk) &&
753 				!ptrace_reparented(tsk) ?
754 			tsk->exit_signal : SIGCHLD;
755 		autoreap = do_notify_parent(tsk, sig);
756 	} else if (thread_group_leader(tsk)) {
757 		autoreap = thread_group_empty(tsk) &&
758 			do_notify_parent(tsk, tsk->exit_signal);
759 	} else {
760 		autoreap = true;
761 	}
762 
763 	if (autoreap) {
764 		tsk->exit_state = EXIT_DEAD;
765 		list_add(&tsk->ptrace_entry, &dead);
766 	}
767 
768 	/* mt-exec, de_thread() is waiting for group leader */
769 	if (unlikely(tsk->signal->notify_count < 0))
770 		wake_up_process(tsk->signal->group_exec_task);
771 	write_unlock_irq(&tasklist_lock);
772 
773 	list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
774 		list_del_init(&p->ptrace_entry);
775 		release_task(p);
776 	}
777 }
778 
779 #ifdef CONFIG_DEBUG_STACK_USAGE
780 unsigned long stack_not_used(struct task_struct *p)
781 {
782 	unsigned long *n = end_of_stack(p);
783 
784 	do {	/* Skip over canary */
785 # ifdef CONFIG_STACK_GROWSUP
786 		n--;
787 # else
788 		n++;
789 # endif
790 	} while (!*n);
791 
792 # ifdef CONFIG_STACK_GROWSUP
793 	return (unsigned long)end_of_stack(p) - (unsigned long)n;
794 # else
795 	return (unsigned long)n - (unsigned long)end_of_stack(p);
796 # endif
797 }
798 
799 /* Count the maximum pages reached in kernel stacks */
800 static inline void kstack_histogram(unsigned long used_stack)
801 {
802 #ifdef CONFIG_VM_EVENT_COUNTERS
803 	if (used_stack <= 1024)
804 		count_vm_event(KSTACK_1K);
805 #if THREAD_SIZE > 1024
806 	else if (used_stack <= 2048)
807 		count_vm_event(KSTACK_2K);
808 #endif
809 #if THREAD_SIZE > 2048
810 	else if (used_stack <= 4096)
811 		count_vm_event(KSTACK_4K);
812 #endif
813 #if THREAD_SIZE > 4096
814 	else if (used_stack <= 8192)
815 		count_vm_event(KSTACK_8K);
816 #endif
817 #if THREAD_SIZE > 8192
818 	else if (used_stack <= 16384)
819 		count_vm_event(KSTACK_16K);
820 #endif
821 #if THREAD_SIZE > 16384
822 	else if (used_stack <= 32768)
823 		count_vm_event(KSTACK_32K);
824 #endif
825 #if THREAD_SIZE > 32768
826 	else if (used_stack <= 65536)
827 		count_vm_event(KSTACK_64K);
828 #endif
829 #if THREAD_SIZE > 65536
830 	else
831 		count_vm_event(KSTACK_REST);
832 #endif
833 #endif /* CONFIG_VM_EVENT_COUNTERS */
834 }
835 
836 static void check_stack_usage(void)
837 {
838 	static DEFINE_SPINLOCK(low_water_lock);
839 	static int lowest_to_date = THREAD_SIZE;
840 	unsigned long free;
841 
842 	free = stack_not_used(current);
843 	kstack_histogram(THREAD_SIZE - free);
844 
845 	if (free >= lowest_to_date)
846 		return;
847 
848 	spin_lock(&low_water_lock);
849 	if (free < lowest_to_date) {
850 		pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
851 			current->comm, task_pid_nr(current), free);
852 		lowest_to_date = free;
853 	}
854 	spin_unlock(&low_water_lock);
855 }
856 #else
857 static inline void check_stack_usage(void) {}
858 #endif
859 
860 static void synchronize_group_exit(struct task_struct *tsk, long code)
861 {
862 	struct sighand_struct *sighand = tsk->sighand;
863 	struct signal_struct *signal = tsk->signal;
864 
865 	spin_lock_irq(&sighand->siglock);
866 	signal->quick_threads--;
867 	if ((signal->quick_threads == 0) &&
868 	    !(signal->flags & SIGNAL_GROUP_EXIT)) {
869 		signal->flags = SIGNAL_GROUP_EXIT;
870 		signal->group_exit_code = code;
871 		signal->group_stop_count = 0;
872 	}
873 	spin_unlock_irq(&sighand->siglock);
874 }
875 
876 void __noreturn do_exit(long code)
877 {
878 	struct task_struct *tsk = current;
879 	int group_dead;
880 
881 	WARN_ON(irqs_disabled());
882 
883 	synchronize_group_exit(tsk, code);
884 
885 	WARN_ON(tsk->plug);
886 
887 	kcov_task_exit(tsk);
888 	kmsan_task_exit(tsk);
889 
890 	coredump_task_exit(tsk);
891 	ptrace_event(PTRACE_EVENT_EXIT, code);
892 	user_events_exit(tsk);
893 
894 	io_uring_files_cancel();
895 	exit_signals(tsk);  /* sets PF_EXITING */
896 
897 	seccomp_filter_release(tsk);
898 
899 	acct_update_integrals(tsk);
900 	group_dead = atomic_dec_and_test(&tsk->signal->live);
901 	if (group_dead) {
902 		/*
903 		 * If the last thread of global init has exited, panic
904 		 * immediately to get a useable coredump.
905 		 */
906 		if (unlikely(is_global_init(tsk)))
907 			panic("Attempted to kill init! exitcode=0x%08x\n",
908 				tsk->signal->group_exit_code ?: (int)code);
909 
910 #ifdef CONFIG_POSIX_TIMERS
911 		hrtimer_cancel(&tsk->signal->real_timer);
912 		exit_itimers(tsk);
913 #endif
914 		if (tsk->mm)
915 			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
916 	}
917 	acct_collect(code, group_dead);
918 	if (group_dead)
919 		tty_audit_exit();
920 	audit_free(tsk);
921 
922 	tsk->exit_code = code;
923 	taskstats_exit(tsk, group_dead);
924 
925 	exit_mm();
926 
927 	if (group_dead)
928 		acct_process();
929 	trace_sched_process_exit(tsk);
930 
931 	exit_sem(tsk);
932 	exit_shm(tsk);
933 	exit_files(tsk);
934 	exit_fs(tsk);
935 	if (group_dead)
936 		disassociate_ctty(1);
937 	exit_task_namespaces(tsk);
938 	exit_task_work(tsk);
939 	exit_thread(tsk);
940 
941 	/*
942 	 * Flush inherited counters to the parent - before the parent
943 	 * gets woken up by child-exit notifications.
944 	 *
945 	 * because of cgroup mode, must be called before cgroup_exit()
946 	 */
947 	perf_event_exit_task(tsk);
948 
949 	sched_autogroup_exit_task(tsk);
950 	cgroup_exit(tsk);
951 
952 	/*
953 	 * FIXME: do that only when needed, using sched_exit tracepoint
954 	 */
955 	flush_ptrace_hw_breakpoint(tsk);
956 
957 	exit_tasks_rcu_start();
958 	exit_notify(tsk, group_dead);
959 	proc_exit_connector(tsk);
960 	mpol_put_task_policy(tsk);
961 #ifdef CONFIG_FUTEX
962 	if (unlikely(current->pi_state_cache))
963 		kfree(current->pi_state_cache);
964 #endif
965 	/*
966 	 * Make sure we are holding no locks:
967 	 */
968 	debug_check_no_locks_held();
969 
970 	if (tsk->io_context)
971 		exit_io_context(tsk);
972 
973 	if (tsk->splice_pipe)
974 		free_pipe_info(tsk->splice_pipe);
975 
976 	if (tsk->task_frag.page)
977 		put_page(tsk->task_frag.page);
978 
979 	exit_task_stack_account(tsk);
980 
981 	check_stack_usage();
982 	preempt_disable();
983 	if (tsk->nr_dirtied)
984 		__this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
985 	exit_rcu();
986 	exit_tasks_rcu_finish();
987 
988 	lockdep_free_task(tsk);
989 	do_task_dead();
990 }
991 
992 void __noreturn make_task_dead(int signr)
993 {
994 	/*
995 	 * Take the task off the cpu after something catastrophic has
996 	 * happened.
997 	 *
998 	 * We can get here from a kernel oops, sometimes with preemption off.
999 	 * Start by checking for critical errors.
1000 	 * Then fix up important state like USER_DS and preemption.
1001 	 * Then do everything else.
1002 	 */
1003 	struct task_struct *tsk = current;
1004 	unsigned int limit;
1005 
1006 	if (unlikely(in_interrupt()))
1007 		panic("Aiee, killing interrupt handler!");
1008 	if (unlikely(!tsk->pid))
1009 		panic("Attempted to kill the idle task!");
1010 
1011 	if (unlikely(irqs_disabled())) {
1012 		pr_info("note: %s[%d] exited with irqs disabled\n",
1013 			current->comm, task_pid_nr(current));
1014 		local_irq_enable();
1015 	}
1016 	if (unlikely(in_atomic())) {
1017 		pr_info("note: %s[%d] exited with preempt_count %d\n",
1018 			current->comm, task_pid_nr(current),
1019 			preempt_count());
1020 		preempt_count_set(PREEMPT_ENABLED);
1021 	}
1022 
1023 	/*
1024 	 * Every time the system oopses, if the oops happens while a reference
1025 	 * to an object was held, the reference leaks.
1026 	 * If the oops doesn't also leak memory, repeated oopsing can cause
1027 	 * reference counters to wrap around (if they're not using refcount_t).
1028 	 * This means that repeated oopsing can make unexploitable-looking bugs
1029 	 * exploitable through repeated oopsing.
1030 	 * To make sure this can't happen, place an upper bound on how often the
1031 	 * kernel may oops without panic().
1032 	 */
1033 	limit = READ_ONCE(oops_limit);
1034 	if (atomic_inc_return(&oops_count) >= limit && limit)
1035 		panic("Oopsed too often (kernel.oops_limit is %d)", limit);
1036 
1037 	/*
1038 	 * We're taking recursive faults here in make_task_dead. Safest is to just
1039 	 * leave this task alone and wait for reboot.
1040 	 */
1041 	if (unlikely(tsk->flags & PF_EXITING)) {
1042 		pr_alert("Fixing recursive fault but reboot is needed!\n");
1043 		futex_exit_recursive(tsk);
1044 		tsk->exit_state = EXIT_DEAD;
1045 		refcount_inc(&tsk->rcu_users);
1046 		do_task_dead();
1047 	}
1048 
1049 	do_exit(signr);
1050 }
1051 
1052 SYSCALL_DEFINE1(exit, int, error_code)
1053 {
1054 	do_exit((error_code&0xff)<<8);
1055 }
1056 
1057 /*
1058  * Take down every thread in the group.  This is called by fatal signals
1059  * as well as by sys_exit_group (below).
1060  */
1061 void __noreturn
1062 do_group_exit(int exit_code)
1063 {
1064 	struct signal_struct *sig = current->signal;
1065 
1066 	if (sig->flags & SIGNAL_GROUP_EXIT)
1067 		exit_code = sig->group_exit_code;
1068 	else if (sig->group_exec_task)
1069 		exit_code = 0;
1070 	else {
1071 		struct sighand_struct *const sighand = current->sighand;
1072 
1073 		spin_lock_irq(&sighand->siglock);
1074 		if (sig->flags & SIGNAL_GROUP_EXIT)
1075 			/* Another thread got here before we took the lock.  */
1076 			exit_code = sig->group_exit_code;
1077 		else if (sig->group_exec_task)
1078 			exit_code = 0;
1079 		else {
1080 			sig->group_exit_code = exit_code;
1081 			sig->flags = SIGNAL_GROUP_EXIT;
1082 			zap_other_threads(current);
1083 		}
1084 		spin_unlock_irq(&sighand->siglock);
1085 	}
1086 
1087 	do_exit(exit_code);
1088 	/* NOTREACHED */
1089 }
1090 
1091 /*
1092  * this kills every thread in the thread group. Note that any externally
1093  * wait4()-ing process will get the correct exit code - even if this
1094  * thread is not the thread group leader.
1095  */
1096 SYSCALL_DEFINE1(exit_group, int, error_code)
1097 {
1098 	do_group_exit((error_code & 0xff) << 8);
1099 	/* NOTREACHED */
1100 	return 0;
1101 }
1102 
1103 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1104 {
1105 	return	wo->wo_type == PIDTYPE_MAX ||
1106 		task_pid_type(p, wo->wo_type) == wo->wo_pid;
1107 }
1108 
1109 static int
1110 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1111 {
1112 	if (!eligible_pid(wo, p))
1113 		return 0;
1114 
1115 	/*
1116 	 * Wait for all children (clone and not) if __WALL is set or
1117 	 * if it is traced by us.
1118 	 */
1119 	if (ptrace || (wo->wo_flags & __WALL))
1120 		return 1;
1121 
1122 	/*
1123 	 * Otherwise, wait for clone children *only* if __WCLONE is set;
1124 	 * otherwise, wait for non-clone children *only*.
1125 	 *
1126 	 * Note: a "clone" child here is one that reports to its parent
1127 	 * using a signal other than SIGCHLD, or a non-leader thread which
1128 	 * we can only see if it is traced by us.
1129 	 */
1130 	if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1131 		return 0;
1132 
1133 	return 1;
1134 }
1135 
1136 /*
1137  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1138  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1139  * the lock and this task is uninteresting.  If we return nonzero, we have
1140  * released the lock and the system call should return.
1141  */
1142 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1143 {
1144 	int state, status;
1145 	pid_t pid = task_pid_vnr(p);
1146 	uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1147 	struct waitid_info *infop;
1148 
1149 	if (!likely(wo->wo_flags & WEXITED))
1150 		return 0;
1151 
1152 	if (unlikely(wo->wo_flags & WNOWAIT)) {
1153 		status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1154 			? p->signal->group_exit_code : p->exit_code;
1155 		get_task_struct(p);
1156 		read_unlock(&tasklist_lock);
1157 		sched_annotate_sleep();
1158 		if (wo->wo_rusage)
1159 			getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1160 		put_task_struct(p);
1161 		goto out_info;
1162 	}
1163 	/*
1164 	 * Move the task's state to DEAD/TRACE, only one thread can do this.
1165 	 */
1166 	state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1167 		EXIT_TRACE : EXIT_DEAD;
1168 	if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1169 		return 0;
1170 	/*
1171 	 * We own this thread, nobody else can reap it.
1172 	 */
1173 	read_unlock(&tasklist_lock);
1174 	sched_annotate_sleep();
1175 
1176 	/*
1177 	 * Check thread_group_leader() to exclude the traced sub-threads.
1178 	 */
1179 	if (state == EXIT_DEAD && thread_group_leader(p)) {
1180 		struct signal_struct *sig = p->signal;
1181 		struct signal_struct *psig = current->signal;
1182 		unsigned long maxrss;
1183 		u64 tgutime, tgstime;
1184 
1185 		/*
1186 		 * The resource counters for the group leader are in its
1187 		 * own task_struct.  Those for dead threads in the group
1188 		 * are in its signal_struct, as are those for the child
1189 		 * processes it has previously reaped.  All these
1190 		 * accumulate in the parent's signal_struct c* fields.
1191 		 *
1192 		 * We don't bother to take a lock here to protect these
1193 		 * p->signal fields because the whole thread group is dead
1194 		 * and nobody can change them.
1195 		 *
1196 		 * psig->stats_lock also protects us from our sub-threads
1197 		 * which can reap other children at the same time.
1198 		 *
1199 		 * We use thread_group_cputime_adjusted() to get times for
1200 		 * the thread group, which consolidates times for all threads
1201 		 * in the group including the group leader.
1202 		 */
1203 		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1204 		write_seqlock_irq(&psig->stats_lock);
1205 		psig->cutime += tgutime + sig->cutime;
1206 		psig->cstime += tgstime + sig->cstime;
1207 		psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1208 		psig->cmin_flt +=
1209 			p->min_flt + sig->min_flt + sig->cmin_flt;
1210 		psig->cmaj_flt +=
1211 			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1212 		psig->cnvcsw +=
1213 			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1214 		psig->cnivcsw +=
1215 			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1216 		psig->cinblock +=
1217 			task_io_get_inblock(p) +
1218 			sig->inblock + sig->cinblock;
1219 		psig->coublock +=
1220 			task_io_get_oublock(p) +
1221 			sig->oublock + sig->coublock;
1222 		maxrss = max(sig->maxrss, sig->cmaxrss);
1223 		if (psig->cmaxrss < maxrss)
1224 			psig->cmaxrss = maxrss;
1225 		task_io_accounting_add(&psig->ioac, &p->ioac);
1226 		task_io_accounting_add(&psig->ioac, &sig->ioac);
1227 		write_sequnlock_irq(&psig->stats_lock);
1228 	}
1229 
1230 	if (wo->wo_rusage)
1231 		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1232 	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1233 		? p->signal->group_exit_code : p->exit_code;
1234 	wo->wo_stat = status;
1235 
1236 	if (state == EXIT_TRACE) {
1237 		write_lock_irq(&tasklist_lock);
1238 		/* We dropped tasklist, ptracer could die and untrace */
1239 		ptrace_unlink(p);
1240 
1241 		/* If parent wants a zombie, don't release it now */
1242 		state = EXIT_ZOMBIE;
1243 		if (do_notify_parent(p, p->exit_signal))
1244 			state = EXIT_DEAD;
1245 		p->exit_state = state;
1246 		write_unlock_irq(&tasklist_lock);
1247 	}
1248 	if (state == EXIT_DEAD)
1249 		release_task(p);
1250 
1251 out_info:
1252 	infop = wo->wo_info;
1253 	if (infop) {
1254 		if ((status & 0x7f) == 0) {
1255 			infop->cause = CLD_EXITED;
1256 			infop->status = status >> 8;
1257 		} else {
1258 			infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1259 			infop->status = status & 0x7f;
1260 		}
1261 		infop->pid = pid;
1262 		infop->uid = uid;
1263 	}
1264 
1265 	return pid;
1266 }
1267 
1268 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1269 {
1270 	if (ptrace) {
1271 		if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1272 			return &p->exit_code;
1273 	} else {
1274 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1275 			return &p->signal->group_exit_code;
1276 	}
1277 	return NULL;
1278 }
1279 
1280 /**
1281  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1282  * @wo: wait options
1283  * @ptrace: is the wait for ptrace
1284  * @p: task to wait for
1285  *
1286  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1287  *
1288  * CONTEXT:
1289  * read_lock(&tasklist_lock), which is released if return value is
1290  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1291  *
1292  * RETURNS:
1293  * 0 if wait condition didn't exist and search for other wait conditions
1294  * should continue.  Non-zero return, -errno on failure and @p's pid on
1295  * success, implies that tasklist_lock is released and wait condition
1296  * search should terminate.
1297  */
1298 static int wait_task_stopped(struct wait_opts *wo,
1299 				int ptrace, struct task_struct *p)
1300 {
1301 	struct waitid_info *infop;
1302 	int exit_code, *p_code, why;
1303 	uid_t uid = 0; /* unneeded, required by compiler */
1304 	pid_t pid;
1305 
1306 	/*
1307 	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1308 	 */
1309 	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1310 		return 0;
1311 
1312 	if (!task_stopped_code(p, ptrace))
1313 		return 0;
1314 
1315 	exit_code = 0;
1316 	spin_lock_irq(&p->sighand->siglock);
1317 
1318 	p_code = task_stopped_code(p, ptrace);
1319 	if (unlikely(!p_code))
1320 		goto unlock_sig;
1321 
1322 	exit_code = *p_code;
1323 	if (!exit_code)
1324 		goto unlock_sig;
1325 
1326 	if (!unlikely(wo->wo_flags & WNOWAIT))
1327 		*p_code = 0;
1328 
1329 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1330 unlock_sig:
1331 	spin_unlock_irq(&p->sighand->siglock);
1332 	if (!exit_code)
1333 		return 0;
1334 
1335 	/*
1336 	 * Now we are pretty sure this task is interesting.
1337 	 * Make sure it doesn't get reaped out from under us while we
1338 	 * give up the lock and then examine it below.  We don't want to
1339 	 * keep holding onto the tasklist_lock while we call getrusage and
1340 	 * possibly take page faults for user memory.
1341 	 */
1342 	get_task_struct(p);
1343 	pid = task_pid_vnr(p);
1344 	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1345 	read_unlock(&tasklist_lock);
1346 	sched_annotate_sleep();
1347 	if (wo->wo_rusage)
1348 		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1349 	put_task_struct(p);
1350 
1351 	if (likely(!(wo->wo_flags & WNOWAIT)))
1352 		wo->wo_stat = (exit_code << 8) | 0x7f;
1353 
1354 	infop = wo->wo_info;
1355 	if (infop) {
1356 		infop->cause = why;
1357 		infop->status = exit_code;
1358 		infop->pid = pid;
1359 		infop->uid = uid;
1360 	}
1361 	return pid;
1362 }
1363 
1364 /*
1365  * Handle do_wait work for one task in a live, non-stopped state.
1366  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1367  * the lock and this task is uninteresting.  If we return nonzero, we have
1368  * released the lock and the system call should return.
1369  */
1370 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1371 {
1372 	struct waitid_info *infop;
1373 	pid_t pid;
1374 	uid_t uid;
1375 
1376 	if (!unlikely(wo->wo_flags & WCONTINUED))
1377 		return 0;
1378 
1379 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1380 		return 0;
1381 
1382 	spin_lock_irq(&p->sighand->siglock);
1383 	/* Re-check with the lock held.  */
1384 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1385 		spin_unlock_irq(&p->sighand->siglock);
1386 		return 0;
1387 	}
1388 	if (!unlikely(wo->wo_flags & WNOWAIT))
1389 		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1390 	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1391 	spin_unlock_irq(&p->sighand->siglock);
1392 
1393 	pid = task_pid_vnr(p);
1394 	get_task_struct(p);
1395 	read_unlock(&tasklist_lock);
1396 	sched_annotate_sleep();
1397 	if (wo->wo_rusage)
1398 		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1399 	put_task_struct(p);
1400 
1401 	infop = wo->wo_info;
1402 	if (!infop) {
1403 		wo->wo_stat = 0xffff;
1404 	} else {
1405 		infop->cause = CLD_CONTINUED;
1406 		infop->pid = pid;
1407 		infop->uid = uid;
1408 		infop->status = SIGCONT;
1409 	}
1410 	return pid;
1411 }
1412 
1413 /*
1414  * Consider @p for a wait by @parent.
1415  *
1416  * -ECHILD should be in ->notask_error before the first call.
1417  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1418  * Returns zero if the search for a child should continue;
1419  * then ->notask_error is 0 if @p is an eligible child,
1420  * or still -ECHILD.
1421  */
1422 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1423 				struct task_struct *p)
1424 {
1425 	/*
1426 	 * We can race with wait_task_zombie() from another thread.
1427 	 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1428 	 * can't confuse the checks below.
1429 	 */
1430 	int exit_state = READ_ONCE(p->exit_state);
1431 	int ret;
1432 
1433 	if (unlikely(exit_state == EXIT_DEAD))
1434 		return 0;
1435 
1436 	ret = eligible_child(wo, ptrace, p);
1437 	if (!ret)
1438 		return ret;
1439 
1440 	if (unlikely(exit_state == EXIT_TRACE)) {
1441 		/*
1442 		 * ptrace == 0 means we are the natural parent. In this case
1443 		 * we should clear notask_error, debugger will notify us.
1444 		 */
1445 		if (likely(!ptrace))
1446 			wo->notask_error = 0;
1447 		return 0;
1448 	}
1449 
1450 	if (likely(!ptrace) && unlikely(p->ptrace)) {
1451 		/*
1452 		 * If it is traced by its real parent's group, just pretend
1453 		 * the caller is ptrace_do_wait() and reap this child if it
1454 		 * is zombie.
1455 		 *
1456 		 * This also hides group stop state from real parent; otherwise
1457 		 * a single stop can be reported twice as group and ptrace stop.
1458 		 * If a ptracer wants to distinguish these two events for its
1459 		 * own children it should create a separate process which takes
1460 		 * the role of real parent.
1461 		 */
1462 		if (!ptrace_reparented(p))
1463 			ptrace = 1;
1464 	}
1465 
1466 	/* slay zombie? */
1467 	if (exit_state == EXIT_ZOMBIE) {
1468 		/* we don't reap group leaders with subthreads */
1469 		if (!delay_group_leader(p)) {
1470 			/*
1471 			 * A zombie ptracee is only visible to its ptracer.
1472 			 * Notification and reaping will be cascaded to the
1473 			 * real parent when the ptracer detaches.
1474 			 */
1475 			if (unlikely(ptrace) || likely(!p->ptrace))
1476 				return wait_task_zombie(wo, p);
1477 		}
1478 
1479 		/*
1480 		 * Allow access to stopped/continued state via zombie by
1481 		 * falling through.  Clearing of notask_error is complex.
1482 		 *
1483 		 * When !@ptrace:
1484 		 *
1485 		 * If WEXITED is set, notask_error should naturally be
1486 		 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1487 		 * so, if there are live subthreads, there are events to
1488 		 * wait for.  If all subthreads are dead, it's still safe
1489 		 * to clear - this function will be called again in finite
1490 		 * amount time once all the subthreads are released and
1491 		 * will then return without clearing.
1492 		 *
1493 		 * When @ptrace:
1494 		 *
1495 		 * Stopped state is per-task and thus can't change once the
1496 		 * target task dies.  Only continued and exited can happen.
1497 		 * Clear notask_error if WCONTINUED | WEXITED.
1498 		 */
1499 		if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1500 			wo->notask_error = 0;
1501 	} else {
1502 		/*
1503 		 * @p is alive and it's gonna stop, continue or exit, so
1504 		 * there always is something to wait for.
1505 		 */
1506 		wo->notask_error = 0;
1507 	}
1508 
1509 	/*
1510 	 * Wait for stopped.  Depending on @ptrace, different stopped state
1511 	 * is used and the two don't interact with each other.
1512 	 */
1513 	ret = wait_task_stopped(wo, ptrace, p);
1514 	if (ret)
1515 		return ret;
1516 
1517 	/*
1518 	 * Wait for continued.  There's only one continued state and the
1519 	 * ptracer can consume it which can confuse the real parent.  Don't
1520 	 * use WCONTINUED from ptracer.  You don't need or want it.
1521 	 */
1522 	return wait_task_continued(wo, p);
1523 }
1524 
1525 /*
1526  * Do the work of do_wait() for one thread in the group, @tsk.
1527  *
1528  * -ECHILD should be in ->notask_error before the first call.
1529  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1530  * Returns zero if the search for a child should continue; then
1531  * ->notask_error is 0 if there were any eligible children,
1532  * or still -ECHILD.
1533  */
1534 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1535 {
1536 	struct task_struct *p;
1537 
1538 	list_for_each_entry(p, &tsk->children, sibling) {
1539 		int ret = wait_consider_task(wo, 0, p);
1540 
1541 		if (ret)
1542 			return ret;
1543 	}
1544 
1545 	return 0;
1546 }
1547 
1548 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1549 {
1550 	struct task_struct *p;
1551 
1552 	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1553 		int ret = wait_consider_task(wo, 1, p);
1554 
1555 		if (ret)
1556 			return ret;
1557 	}
1558 
1559 	return 0;
1560 }
1561 
1562 bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1563 {
1564 	if (!eligible_pid(wo, p))
1565 		return false;
1566 
1567 	if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1568 		return false;
1569 
1570 	return true;
1571 }
1572 
1573 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1574 				int sync, void *key)
1575 {
1576 	struct wait_opts *wo = container_of(wait, struct wait_opts,
1577 						child_wait);
1578 	struct task_struct *p = key;
1579 
1580 	if (pid_child_should_wake(wo, p))
1581 		return default_wake_function(wait, mode, sync, key);
1582 
1583 	return 0;
1584 }
1585 
1586 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1587 {
1588 	__wake_up_sync_key(&parent->signal->wait_chldexit,
1589 			   TASK_INTERRUPTIBLE, p);
1590 }
1591 
1592 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1593 				 struct task_struct *target)
1594 {
1595 	struct task_struct *parent =
1596 		!ptrace ? target->real_parent : target->parent;
1597 
1598 	return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1599 				     same_thread_group(current, parent));
1600 }
1601 
1602 /*
1603  * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1604  * and tracee lists to find the target task.
1605  */
1606 static int do_wait_pid(struct wait_opts *wo)
1607 {
1608 	bool ptrace;
1609 	struct task_struct *target;
1610 	int retval;
1611 
1612 	ptrace = false;
1613 	target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1614 	if (target && is_effectively_child(wo, ptrace, target)) {
1615 		retval = wait_consider_task(wo, ptrace, target);
1616 		if (retval)
1617 			return retval;
1618 	}
1619 
1620 	ptrace = true;
1621 	target = pid_task(wo->wo_pid, PIDTYPE_PID);
1622 	if (target && target->ptrace &&
1623 	    is_effectively_child(wo, ptrace, target)) {
1624 		retval = wait_consider_task(wo, ptrace, target);
1625 		if (retval)
1626 			return retval;
1627 	}
1628 
1629 	return 0;
1630 }
1631 
1632 long __do_wait(struct wait_opts *wo)
1633 {
1634 	long retval;
1635 
1636 	/*
1637 	 * If there is nothing that can match our criteria, just get out.
1638 	 * We will clear ->notask_error to zero if we see any child that
1639 	 * might later match our criteria, even if we are not able to reap
1640 	 * it yet.
1641 	 */
1642 	wo->notask_error = -ECHILD;
1643 	if ((wo->wo_type < PIDTYPE_MAX) &&
1644 	   (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1645 		goto notask;
1646 
1647 	read_lock(&tasklist_lock);
1648 
1649 	if (wo->wo_type == PIDTYPE_PID) {
1650 		retval = do_wait_pid(wo);
1651 		if (retval)
1652 			return retval;
1653 	} else {
1654 		struct task_struct *tsk = current;
1655 
1656 		do {
1657 			retval = do_wait_thread(wo, tsk);
1658 			if (retval)
1659 				return retval;
1660 
1661 			retval = ptrace_do_wait(wo, tsk);
1662 			if (retval)
1663 				return retval;
1664 
1665 			if (wo->wo_flags & __WNOTHREAD)
1666 				break;
1667 		} while_each_thread(current, tsk);
1668 	}
1669 	read_unlock(&tasklist_lock);
1670 
1671 notask:
1672 	retval = wo->notask_error;
1673 	if (!retval && !(wo->wo_flags & WNOHANG))
1674 		return -ERESTARTSYS;
1675 
1676 	return retval;
1677 }
1678 
1679 static long do_wait(struct wait_opts *wo)
1680 {
1681 	int retval;
1682 
1683 	trace_sched_process_wait(wo->wo_pid);
1684 
1685 	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1686 	wo->child_wait.private = current;
1687 	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1688 
1689 	do {
1690 		set_current_state(TASK_INTERRUPTIBLE);
1691 		retval = __do_wait(wo);
1692 		if (retval != -ERESTARTSYS)
1693 			break;
1694 		if (signal_pending(current))
1695 			break;
1696 		schedule();
1697 	} while (1);
1698 
1699 	__set_current_state(TASK_RUNNING);
1700 	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1701 	return retval;
1702 }
1703 
1704 int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1705 			  struct waitid_info *infop, int options,
1706 			  struct rusage *ru)
1707 {
1708 	unsigned int f_flags = 0;
1709 	struct pid *pid = NULL;
1710 	enum pid_type type;
1711 
1712 	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1713 			__WNOTHREAD|__WCLONE|__WALL))
1714 		return -EINVAL;
1715 	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1716 		return -EINVAL;
1717 
1718 	switch (which) {
1719 	case P_ALL:
1720 		type = PIDTYPE_MAX;
1721 		break;
1722 	case P_PID:
1723 		type = PIDTYPE_PID;
1724 		if (upid <= 0)
1725 			return -EINVAL;
1726 
1727 		pid = find_get_pid(upid);
1728 		break;
1729 	case P_PGID:
1730 		type = PIDTYPE_PGID;
1731 		if (upid < 0)
1732 			return -EINVAL;
1733 
1734 		if (upid)
1735 			pid = find_get_pid(upid);
1736 		else
1737 			pid = get_task_pid(current, PIDTYPE_PGID);
1738 		break;
1739 	case P_PIDFD:
1740 		type = PIDTYPE_PID;
1741 		if (upid < 0)
1742 			return -EINVAL;
1743 
1744 		pid = pidfd_get_pid(upid, &f_flags);
1745 		if (IS_ERR(pid))
1746 			return PTR_ERR(pid);
1747 
1748 		break;
1749 	default:
1750 		return -EINVAL;
1751 	}
1752 
1753 	wo->wo_type	= type;
1754 	wo->wo_pid	= pid;
1755 	wo->wo_flags	= options;
1756 	wo->wo_info	= infop;
1757 	wo->wo_rusage	= ru;
1758 	if (f_flags & O_NONBLOCK)
1759 		wo->wo_flags |= WNOHANG;
1760 
1761 	return 0;
1762 }
1763 
1764 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1765 			  int options, struct rusage *ru)
1766 {
1767 	struct wait_opts wo;
1768 	long ret;
1769 
1770 	ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1771 	if (ret)
1772 		return ret;
1773 
1774 	ret = do_wait(&wo);
1775 	if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1776 		ret = -EAGAIN;
1777 
1778 	put_pid(wo.wo_pid);
1779 	return ret;
1780 }
1781 
1782 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1783 		infop, int, options, struct rusage __user *, ru)
1784 {
1785 	struct rusage r;
1786 	struct waitid_info info = {.status = 0};
1787 	long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1788 	int signo = 0;
1789 
1790 	if (err > 0) {
1791 		signo = SIGCHLD;
1792 		err = 0;
1793 		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1794 			return -EFAULT;
1795 	}
1796 	if (!infop)
1797 		return err;
1798 
1799 	if (!user_write_access_begin(infop, sizeof(*infop)))
1800 		return -EFAULT;
1801 
1802 	unsafe_put_user(signo, &infop->si_signo, Efault);
1803 	unsafe_put_user(0, &infop->si_errno, Efault);
1804 	unsafe_put_user(info.cause, &infop->si_code, Efault);
1805 	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1806 	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1807 	unsafe_put_user(info.status, &infop->si_status, Efault);
1808 	user_write_access_end();
1809 	return err;
1810 Efault:
1811 	user_write_access_end();
1812 	return -EFAULT;
1813 }
1814 
1815 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1816 		  struct rusage *ru)
1817 {
1818 	struct wait_opts wo;
1819 	struct pid *pid = NULL;
1820 	enum pid_type type;
1821 	long ret;
1822 
1823 	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1824 			__WNOTHREAD|__WCLONE|__WALL))
1825 		return -EINVAL;
1826 
1827 	/* -INT_MIN is not defined */
1828 	if (upid == INT_MIN)
1829 		return -ESRCH;
1830 
1831 	if (upid == -1)
1832 		type = PIDTYPE_MAX;
1833 	else if (upid < 0) {
1834 		type = PIDTYPE_PGID;
1835 		pid = find_get_pid(-upid);
1836 	} else if (upid == 0) {
1837 		type = PIDTYPE_PGID;
1838 		pid = get_task_pid(current, PIDTYPE_PGID);
1839 	} else /* upid > 0 */ {
1840 		type = PIDTYPE_PID;
1841 		pid = find_get_pid(upid);
1842 	}
1843 
1844 	wo.wo_type	= type;
1845 	wo.wo_pid	= pid;
1846 	wo.wo_flags	= options | WEXITED;
1847 	wo.wo_info	= NULL;
1848 	wo.wo_stat	= 0;
1849 	wo.wo_rusage	= ru;
1850 	ret = do_wait(&wo);
1851 	put_pid(pid);
1852 	if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1853 		ret = -EFAULT;
1854 
1855 	return ret;
1856 }
1857 
1858 int kernel_wait(pid_t pid, int *stat)
1859 {
1860 	struct wait_opts wo = {
1861 		.wo_type	= PIDTYPE_PID,
1862 		.wo_pid		= find_get_pid(pid),
1863 		.wo_flags	= WEXITED,
1864 	};
1865 	int ret;
1866 
1867 	ret = do_wait(&wo);
1868 	if (ret > 0 && wo.wo_stat)
1869 		*stat = wo.wo_stat;
1870 	put_pid(wo.wo_pid);
1871 	return ret;
1872 }
1873 
1874 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1875 		int, options, struct rusage __user *, ru)
1876 {
1877 	struct rusage r;
1878 	long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1879 
1880 	if (err > 0) {
1881 		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1882 			return -EFAULT;
1883 	}
1884 	return err;
1885 }
1886 
1887 #ifdef __ARCH_WANT_SYS_WAITPID
1888 
1889 /*
1890  * sys_waitpid() remains for compatibility. waitpid() should be
1891  * implemented by calling sys_wait4() from libc.a.
1892  */
1893 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1894 {
1895 	return kernel_wait4(pid, stat_addr, options, NULL);
1896 }
1897 
1898 #endif
1899 
1900 #ifdef CONFIG_COMPAT
1901 COMPAT_SYSCALL_DEFINE4(wait4,
1902 	compat_pid_t, pid,
1903 	compat_uint_t __user *, stat_addr,
1904 	int, options,
1905 	struct compat_rusage __user *, ru)
1906 {
1907 	struct rusage r;
1908 	long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1909 	if (err > 0) {
1910 		if (ru && put_compat_rusage(&r, ru))
1911 			return -EFAULT;
1912 	}
1913 	return err;
1914 }
1915 
1916 COMPAT_SYSCALL_DEFINE5(waitid,
1917 		int, which, compat_pid_t, pid,
1918 		struct compat_siginfo __user *, infop, int, options,
1919 		struct compat_rusage __user *, uru)
1920 {
1921 	struct rusage ru;
1922 	struct waitid_info info = {.status = 0};
1923 	long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1924 	int signo = 0;
1925 	if (err > 0) {
1926 		signo = SIGCHLD;
1927 		err = 0;
1928 		if (uru) {
1929 			/* kernel_waitid() overwrites everything in ru */
1930 			if (COMPAT_USE_64BIT_TIME)
1931 				err = copy_to_user(uru, &ru, sizeof(ru));
1932 			else
1933 				err = put_compat_rusage(&ru, uru);
1934 			if (err)
1935 				return -EFAULT;
1936 		}
1937 	}
1938 
1939 	if (!infop)
1940 		return err;
1941 
1942 	if (!user_write_access_begin(infop, sizeof(*infop)))
1943 		return -EFAULT;
1944 
1945 	unsafe_put_user(signo, &infop->si_signo, Efault);
1946 	unsafe_put_user(0, &infop->si_errno, Efault);
1947 	unsafe_put_user(info.cause, &infop->si_code, Efault);
1948 	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1949 	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1950 	unsafe_put_user(info.status, &infop->si_status, Efault);
1951 	user_write_access_end();
1952 	return err;
1953 Efault:
1954 	user_write_access_end();
1955 	return -EFAULT;
1956 }
1957 #endif
1958 
1959 /*
1960  * This needs to be __function_aligned as GCC implicitly makes any
1961  * implementation of abort() cold and drops alignment specified by
1962  * -falign-functions=N.
1963  *
1964  * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1965  */
1966 __weak __function_aligned void abort(void)
1967 {
1968 	BUG();
1969 
1970 	/* if that doesn't kill us, halt */
1971 	panic("Oops failed to kill thread");
1972 }
1973 EXPORT_SYMBOL(abort);
1974