xref: /openbmc/linux/kernel/exit.c (revision c4a54f70)
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/blkdev.h>
52  #include <linux/task_io_accounting_ops.h>
53  #include <linux/tracehook.h>
54  #include <linux/fs_struct.h>
55  #include <linux/init_task.h>
56  #include <linux/perf_event.h>
57  #include <trace/events/sched.h>
58  #include <linux/hw_breakpoint.h>
59  #include <linux/oom.h>
60  #include <linux/writeback.h>
61  #include <linux/shm.h>
62  #include <linux/kcov.h>
63  #include <linux/random.h>
64  #include <linux/rcuwait.h>
65  #include <linux/compat.h>
66  
67  #include <linux/uaccess.h>
68  #include <asm/unistd.h>
69  #include <asm/mmu_context.h>
70  
71  static void __unhash_process(struct task_struct *p, bool group_dead)
72  {
73  	nr_threads--;
74  	detach_pid(p, PIDTYPE_PID);
75  	if (group_dead) {
76  		detach_pid(p, PIDTYPE_TGID);
77  		detach_pid(p, PIDTYPE_PGID);
78  		detach_pid(p, PIDTYPE_SID);
79  
80  		list_del_rcu(&p->tasks);
81  		list_del_init(&p->sibling);
82  		__this_cpu_dec(process_counts);
83  	}
84  	list_del_rcu(&p->thread_group);
85  	list_del_rcu(&p->thread_node);
86  }
87  
88  /*
89   * This function expects the tasklist_lock write-locked.
90   */
91  static void __exit_signal(struct task_struct *tsk)
92  {
93  	struct signal_struct *sig = tsk->signal;
94  	bool group_dead = thread_group_leader(tsk);
95  	struct sighand_struct *sighand;
96  	struct tty_struct *uninitialized_var(tty);
97  	u64 utime, stime;
98  
99  	sighand = rcu_dereference_check(tsk->sighand,
100  					lockdep_tasklist_lock_is_held());
101  	spin_lock(&sighand->siglock);
102  
103  #ifdef CONFIG_POSIX_TIMERS
104  	posix_cpu_timers_exit(tsk);
105  	if (group_dead)
106  		posix_cpu_timers_exit_group(tsk);
107  #endif
108  
109  	if (group_dead) {
110  		tty = sig->tty;
111  		sig->tty = NULL;
112  	} else {
113  		/*
114  		 * If there is any task waiting for the group exit
115  		 * then notify it:
116  		 */
117  		if (sig->notify_count > 0 && !--sig->notify_count)
118  			wake_up_process(sig->group_exit_task);
119  
120  		if (tsk == sig->curr_target)
121  			sig->curr_target = next_thread(tsk);
122  	}
123  
124  	add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
125  			      sizeof(unsigned long long));
126  
127  	/*
128  	 * Accumulate here the counters for all threads as they die. We could
129  	 * skip the group leader because it is the last user of signal_struct,
130  	 * but we want to avoid the race with thread_group_cputime() which can
131  	 * see the empty ->thread_head list.
132  	 */
133  	task_cputime(tsk, &utime, &stime);
134  	write_seqlock(&sig->stats_lock);
135  	sig->utime += utime;
136  	sig->stime += stime;
137  	sig->gtime += task_gtime(tsk);
138  	sig->min_flt += tsk->min_flt;
139  	sig->maj_flt += tsk->maj_flt;
140  	sig->nvcsw += tsk->nvcsw;
141  	sig->nivcsw += tsk->nivcsw;
142  	sig->inblock += task_io_get_inblock(tsk);
143  	sig->oublock += task_io_get_oublock(tsk);
144  	task_io_accounting_add(&sig->ioac, &tsk->ioac);
145  	sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
146  	sig->nr_threads--;
147  	__unhash_process(tsk, group_dead);
148  	write_sequnlock(&sig->stats_lock);
149  
150  	/*
151  	 * Do this under ->siglock, we can race with another thread
152  	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
153  	 */
154  	flush_sigqueue(&tsk->pending);
155  	tsk->sighand = NULL;
156  	spin_unlock(&sighand->siglock);
157  
158  	__cleanup_sighand(sighand);
159  	clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
160  	if (group_dead) {
161  		flush_sigqueue(&sig->shared_pending);
162  		tty_kref_put(tty);
163  	}
164  }
165  
166  static void delayed_put_task_struct(struct rcu_head *rhp)
167  {
168  	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
169  
170  	perf_event_delayed_put(tsk);
171  	trace_sched_process_free(tsk);
172  	put_task_struct(tsk);
173  }
174  
175  void put_task_struct_rcu_user(struct task_struct *task)
176  {
177  	if (refcount_dec_and_test(&task->rcu_users))
178  		call_rcu(&task->rcu, delayed_put_task_struct);
179  }
180  
181  void release_task(struct task_struct *p)
182  {
183  	struct task_struct *leader;
184  	struct pid *thread_pid;
185  	int zap_leader;
186  repeat:
187  	/* don't need to get the RCU readlock here - the process is dead and
188  	 * can't be modifying its own credentials. But shut RCU-lockdep up */
189  	rcu_read_lock();
190  	atomic_dec(&__task_cred(p)->user->processes);
191  	rcu_read_unlock();
192  
193  	cgroup_release(p);
194  
195  	write_lock_irq(&tasklist_lock);
196  	ptrace_release_task(p);
197  	thread_pid = get_pid(p->thread_pid);
198  	__exit_signal(p);
199  
200  	/*
201  	 * If we are the last non-leader member of the thread
202  	 * group, and the leader is zombie, then notify the
203  	 * group leader's parent process. (if it wants notification.)
204  	 */
205  	zap_leader = 0;
206  	leader = p->group_leader;
207  	if (leader != p && thread_group_empty(leader)
208  			&& leader->exit_state == EXIT_ZOMBIE) {
209  		/*
210  		 * If we were the last child thread and the leader has
211  		 * exited already, and the leader's parent ignores SIGCHLD,
212  		 * then we are the one who should release the leader.
213  		 */
214  		zap_leader = do_notify_parent(leader, leader->exit_signal);
215  		if (zap_leader)
216  			leader->exit_state = EXIT_DEAD;
217  	}
218  
219  	write_unlock_irq(&tasklist_lock);
220  	proc_flush_pid(thread_pid);
221  	put_pid(thread_pid);
222  	release_thread(p);
223  	put_task_struct_rcu_user(p);
224  
225  	p = leader;
226  	if (unlikely(zap_leader))
227  		goto repeat;
228  }
229  
230  int rcuwait_wake_up(struct rcuwait *w)
231  {
232  	int ret = 0;
233  	struct task_struct *task;
234  
235  	rcu_read_lock();
236  
237  	/*
238  	 * Order condition vs @task, such that everything prior to the load
239  	 * of @task is visible. This is the condition as to why the user called
240  	 * rcuwait_wake() in the first place. Pairs with set_current_state()
241  	 * barrier (A) in rcuwait_wait_event().
242  	 *
243  	 *    WAIT                WAKE
244  	 *    [S] tsk = current	  [S] cond = true
245  	 *        MB (A)	      MB (B)
246  	 *    [L] cond		  [L] tsk
247  	 */
248  	smp_mb(); /* (B) */
249  
250  	task = rcu_dereference(w->task);
251  	if (task)
252  		ret = wake_up_process(task);
253  	rcu_read_unlock();
254  
255  	return ret;
256  }
257  EXPORT_SYMBOL_GPL(rcuwait_wake_up);
258  
259  /*
260   * Determine if a process group is "orphaned", according to the POSIX
261   * definition in 2.2.2.52.  Orphaned process groups are not to be affected
262   * by terminal-generated stop signals.  Newly orphaned process groups are
263   * to receive a SIGHUP and a SIGCONT.
264   *
265   * "I ask you, have you ever known what it is to be an orphan?"
266   */
267  static int will_become_orphaned_pgrp(struct pid *pgrp,
268  					struct task_struct *ignored_task)
269  {
270  	struct task_struct *p;
271  
272  	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
273  		if ((p == ignored_task) ||
274  		    (p->exit_state && thread_group_empty(p)) ||
275  		    is_global_init(p->real_parent))
276  			continue;
277  
278  		if (task_pgrp(p->real_parent) != pgrp &&
279  		    task_session(p->real_parent) == task_session(p))
280  			return 0;
281  	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
282  
283  	return 1;
284  }
285  
286  int is_current_pgrp_orphaned(void)
287  {
288  	int retval;
289  
290  	read_lock(&tasklist_lock);
291  	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
292  	read_unlock(&tasklist_lock);
293  
294  	return retval;
295  }
296  
297  static bool has_stopped_jobs(struct pid *pgrp)
298  {
299  	struct task_struct *p;
300  
301  	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
302  		if (p->signal->flags & SIGNAL_STOP_STOPPED)
303  			return true;
304  	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
305  
306  	return false;
307  }
308  
309  /*
310   * Check to see if any process groups have become orphaned as
311   * a result of our exiting, and if they have any stopped jobs,
312   * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
313   */
314  static void
315  kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
316  {
317  	struct pid *pgrp = task_pgrp(tsk);
318  	struct task_struct *ignored_task = tsk;
319  
320  	if (!parent)
321  		/* exit: our father is in a different pgrp than
322  		 * we are and we were the only connection outside.
323  		 */
324  		parent = tsk->real_parent;
325  	else
326  		/* reparent: our child is in a different pgrp than
327  		 * we are, and it was the only connection outside.
328  		 */
329  		ignored_task = NULL;
330  
331  	if (task_pgrp(parent) != pgrp &&
332  	    task_session(parent) == task_session(tsk) &&
333  	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
334  	    has_stopped_jobs(pgrp)) {
335  		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
336  		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
337  	}
338  }
339  
340  #ifdef CONFIG_MEMCG
341  /*
342   * A task is exiting.   If it owned this mm, find a new owner for the mm.
343   */
344  void mm_update_next_owner(struct mm_struct *mm)
345  {
346  	struct task_struct *c, *g, *p = current;
347  
348  retry:
349  	/*
350  	 * If the exiting or execing task is not the owner, it's
351  	 * someone else's problem.
352  	 */
353  	if (mm->owner != p)
354  		return;
355  	/*
356  	 * The current owner is exiting/execing and there are no other
357  	 * candidates.  Do not leave the mm pointing to a possibly
358  	 * freed task structure.
359  	 */
360  	if (atomic_read(&mm->mm_users) <= 1) {
361  		WRITE_ONCE(mm->owner, NULL);
362  		return;
363  	}
364  
365  	read_lock(&tasklist_lock);
366  	/*
367  	 * Search in the children
368  	 */
369  	list_for_each_entry(c, &p->children, sibling) {
370  		if (c->mm == mm)
371  			goto assign_new_owner;
372  	}
373  
374  	/*
375  	 * Search in the siblings
376  	 */
377  	list_for_each_entry(c, &p->real_parent->children, sibling) {
378  		if (c->mm == mm)
379  			goto assign_new_owner;
380  	}
381  
382  	/*
383  	 * Search through everything else, we should not get here often.
384  	 */
385  	for_each_process(g) {
386  		if (g->flags & PF_KTHREAD)
387  			continue;
388  		for_each_thread(g, c) {
389  			if (c->mm == mm)
390  				goto assign_new_owner;
391  			if (c->mm)
392  				break;
393  		}
394  	}
395  	read_unlock(&tasklist_lock);
396  	/*
397  	 * We found no owner yet mm_users > 1: this implies that we are
398  	 * most likely racing with swapoff (try_to_unuse()) or /proc or
399  	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
400  	 */
401  	WRITE_ONCE(mm->owner, NULL);
402  	return;
403  
404  assign_new_owner:
405  	BUG_ON(c == p);
406  	get_task_struct(c);
407  	/*
408  	 * The task_lock protects c->mm from changing.
409  	 * We always want mm->owner->mm == mm
410  	 */
411  	task_lock(c);
412  	/*
413  	 * Delay read_unlock() till we have the task_lock()
414  	 * to ensure that c does not slip away underneath us
415  	 */
416  	read_unlock(&tasklist_lock);
417  	if (c->mm != mm) {
418  		task_unlock(c);
419  		put_task_struct(c);
420  		goto retry;
421  	}
422  	WRITE_ONCE(mm->owner, c);
423  	task_unlock(c);
424  	put_task_struct(c);
425  }
426  #endif /* CONFIG_MEMCG */
427  
428  /*
429   * Turn us into a lazy TLB process if we
430   * aren't already..
431   */
432  static void exit_mm(void)
433  {
434  	struct mm_struct *mm = current->mm;
435  	struct core_state *core_state;
436  
437  	exit_mm_release(current, mm);
438  	if (!mm)
439  		return;
440  	sync_mm_rss(mm);
441  	/*
442  	 * Serialize with any possible pending coredump.
443  	 * We must hold mmap_lock around checking core_state
444  	 * and clearing tsk->mm.  The core-inducing thread
445  	 * will increment ->nr_threads for each thread in the
446  	 * group with ->mm != NULL.
447  	 */
448  	mmap_read_lock(mm);
449  	core_state = mm->core_state;
450  	if (core_state) {
451  		struct core_thread self;
452  
453  		mmap_read_unlock(mm);
454  
455  		self.task = current;
456  		self.next = xchg(&core_state->dumper.next, &self);
457  		/*
458  		 * Implies mb(), the result of xchg() must be visible
459  		 * to core_state->dumper.
460  		 */
461  		if (atomic_dec_and_test(&core_state->nr_threads))
462  			complete(&core_state->startup);
463  
464  		for (;;) {
465  			set_current_state(TASK_UNINTERRUPTIBLE);
466  			if (!self.task) /* see coredump_finish() */
467  				break;
468  			freezable_schedule();
469  		}
470  		__set_current_state(TASK_RUNNING);
471  		mmap_read_lock(mm);
472  	}
473  	mmgrab(mm);
474  	BUG_ON(mm != current->active_mm);
475  	/* more a memory barrier than a real lock */
476  	task_lock(current);
477  	current->mm = NULL;
478  	mmap_read_unlock(mm);
479  	enter_lazy_tlb(mm, current);
480  	task_unlock(current);
481  	mm_update_next_owner(mm);
482  	mmput(mm);
483  	if (test_thread_flag(TIF_MEMDIE))
484  		exit_oom_victim();
485  }
486  
487  static struct task_struct *find_alive_thread(struct task_struct *p)
488  {
489  	struct task_struct *t;
490  
491  	for_each_thread(p, t) {
492  		if (!(t->flags & PF_EXITING))
493  			return t;
494  	}
495  	return NULL;
496  }
497  
498  static struct task_struct *find_child_reaper(struct task_struct *father,
499  						struct list_head *dead)
500  	__releases(&tasklist_lock)
501  	__acquires(&tasklist_lock)
502  {
503  	struct pid_namespace *pid_ns = task_active_pid_ns(father);
504  	struct task_struct *reaper = pid_ns->child_reaper;
505  	struct task_struct *p, *n;
506  
507  	if (likely(reaper != father))
508  		return reaper;
509  
510  	reaper = find_alive_thread(father);
511  	if (reaper) {
512  		pid_ns->child_reaper = reaper;
513  		return reaper;
514  	}
515  
516  	write_unlock_irq(&tasklist_lock);
517  
518  	list_for_each_entry_safe(p, n, dead, ptrace_entry) {
519  		list_del_init(&p->ptrace_entry);
520  		release_task(p);
521  	}
522  
523  	zap_pid_ns_processes(pid_ns);
524  	write_lock_irq(&tasklist_lock);
525  
526  	return father;
527  }
528  
529  /*
530   * When we die, we re-parent all our children, and try to:
531   * 1. give them to another thread in our thread group, if such a member exists
532   * 2. give it to the first ancestor process which prctl'd itself as a
533   *    child_subreaper for its children (like a service manager)
534   * 3. give it to the init process (PID 1) in our pid namespace
535   */
536  static struct task_struct *find_new_reaper(struct task_struct *father,
537  					   struct task_struct *child_reaper)
538  {
539  	struct task_struct *thread, *reaper;
540  
541  	thread = find_alive_thread(father);
542  	if (thread)
543  		return thread;
544  
545  	if (father->signal->has_child_subreaper) {
546  		unsigned int ns_level = task_pid(father)->level;
547  		/*
548  		 * Find the first ->is_child_subreaper ancestor in our pid_ns.
549  		 * We can't check reaper != child_reaper to ensure we do not
550  		 * cross the namespaces, the exiting parent could be injected
551  		 * by setns() + fork().
552  		 * We check pid->level, this is slightly more efficient than
553  		 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
554  		 */
555  		for (reaper = father->real_parent;
556  		     task_pid(reaper)->level == ns_level;
557  		     reaper = reaper->real_parent) {
558  			if (reaper == &init_task)
559  				break;
560  			if (!reaper->signal->is_child_subreaper)
561  				continue;
562  			thread = find_alive_thread(reaper);
563  			if (thread)
564  				return thread;
565  		}
566  	}
567  
568  	return child_reaper;
569  }
570  
571  /*
572  * Any that need to be release_task'd are put on the @dead list.
573   */
574  static void reparent_leader(struct task_struct *father, struct task_struct *p,
575  				struct list_head *dead)
576  {
577  	if (unlikely(p->exit_state == EXIT_DEAD))
578  		return;
579  
580  	/* We don't want people slaying init. */
581  	p->exit_signal = SIGCHLD;
582  
583  	/* If it has exited notify the new parent about this child's death. */
584  	if (!p->ptrace &&
585  	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
586  		if (do_notify_parent(p, p->exit_signal)) {
587  			p->exit_state = EXIT_DEAD;
588  			list_add(&p->ptrace_entry, dead);
589  		}
590  	}
591  
592  	kill_orphaned_pgrp(p, father);
593  }
594  
595  /*
596   * This does two things:
597   *
598   * A.  Make init inherit all the child processes
599   * B.  Check to see if any process groups have become orphaned
600   *	as a result of our exiting, and if they have any stopped
601   *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
602   */
603  static void forget_original_parent(struct task_struct *father,
604  					struct list_head *dead)
605  {
606  	struct task_struct *p, *t, *reaper;
607  
608  	if (unlikely(!list_empty(&father->ptraced)))
609  		exit_ptrace(father, dead);
610  
611  	/* Can drop and reacquire tasklist_lock */
612  	reaper = find_child_reaper(father, dead);
613  	if (list_empty(&father->children))
614  		return;
615  
616  	reaper = find_new_reaper(father, reaper);
617  	list_for_each_entry(p, &father->children, sibling) {
618  		for_each_thread(p, t) {
619  			RCU_INIT_POINTER(t->real_parent, reaper);
620  			BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
621  			if (likely(!t->ptrace))
622  				t->parent = t->real_parent;
623  			if (t->pdeath_signal)
624  				group_send_sig_info(t->pdeath_signal,
625  						    SEND_SIG_NOINFO, t,
626  						    PIDTYPE_TGID);
627  		}
628  		/*
629  		 * If this is a threaded reparent there is no need to
630  		 * notify anyone anything has happened.
631  		 */
632  		if (!same_thread_group(reaper, father))
633  			reparent_leader(father, p, dead);
634  	}
635  	list_splice_tail_init(&father->children, &reaper->children);
636  }
637  
638  /*
639   * Send signals to all our closest relatives so that they know
640   * to properly mourn us..
641   */
642  static void exit_notify(struct task_struct *tsk, int group_dead)
643  {
644  	bool autoreap;
645  	struct task_struct *p, *n;
646  	LIST_HEAD(dead);
647  
648  	write_lock_irq(&tasklist_lock);
649  	forget_original_parent(tsk, &dead);
650  
651  	if (group_dead)
652  		kill_orphaned_pgrp(tsk->group_leader, NULL);
653  
654  	tsk->exit_state = EXIT_ZOMBIE;
655  	if (unlikely(tsk->ptrace)) {
656  		int sig = thread_group_leader(tsk) &&
657  				thread_group_empty(tsk) &&
658  				!ptrace_reparented(tsk) ?
659  			tsk->exit_signal : SIGCHLD;
660  		autoreap = do_notify_parent(tsk, sig);
661  	} else if (thread_group_leader(tsk)) {
662  		autoreap = thread_group_empty(tsk) &&
663  			do_notify_parent(tsk, tsk->exit_signal);
664  	} else {
665  		autoreap = true;
666  	}
667  
668  	if (autoreap) {
669  		tsk->exit_state = EXIT_DEAD;
670  		list_add(&tsk->ptrace_entry, &dead);
671  	}
672  
673  	/* mt-exec, de_thread() is waiting for group leader */
674  	if (unlikely(tsk->signal->notify_count < 0))
675  		wake_up_process(tsk->signal->group_exit_task);
676  	write_unlock_irq(&tasklist_lock);
677  
678  	list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
679  		list_del_init(&p->ptrace_entry);
680  		release_task(p);
681  	}
682  }
683  
684  #ifdef CONFIG_DEBUG_STACK_USAGE
685  static void check_stack_usage(void)
686  {
687  	static DEFINE_SPINLOCK(low_water_lock);
688  	static int lowest_to_date = THREAD_SIZE;
689  	unsigned long free;
690  
691  	free = stack_not_used(current);
692  
693  	if (free >= lowest_to_date)
694  		return;
695  
696  	spin_lock(&low_water_lock);
697  	if (free < lowest_to_date) {
698  		pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
699  			current->comm, task_pid_nr(current), free);
700  		lowest_to_date = free;
701  	}
702  	spin_unlock(&low_water_lock);
703  }
704  #else
705  static inline void check_stack_usage(void) {}
706  #endif
707  
708  void __noreturn do_exit(long code)
709  {
710  	struct task_struct *tsk = current;
711  	int group_dead;
712  
713  	/*
714  	 * We can get here from a kernel oops, sometimes with preemption off.
715  	 * Start by checking for critical errors.
716  	 * Then fix up important state like USER_DS and preemption.
717  	 * Then do everything else.
718  	 */
719  
720  	WARN_ON(blk_needs_flush_plug(tsk));
721  
722  	if (unlikely(in_interrupt()))
723  		panic("Aiee, killing interrupt handler!");
724  	if (unlikely(!tsk->pid))
725  		panic("Attempted to kill the idle task!");
726  
727  	/*
728  	 * If do_exit is called because this processes oopsed, it's possible
729  	 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
730  	 * continuing. Amongst other possible reasons, this is to prevent
731  	 * mm_release()->clear_child_tid() from writing to a user-controlled
732  	 * kernel address.
733  	 */
734  	set_fs(USER_DS);
735  
736  	if (unlikely(in_atomic())) {
737  		pr_info("note: %s[%d] exited with preempt_count %d\n",
738  			current->comm, task_pid_nr(current),
739  			preempt_count());
740  		preempt_count_set(PREEMPT_ENABLED);
741  	}
742  
743  	profile_task_exit(tsk);
744  	kcov_task_exit(tsk);
745  
746  	ptrace_event(PTRACE_EVENT_EXIT, code);
747  
748  	validate_creds_for_do_exit(tsk);
749  
750  	/*
751  	 * We're taking recursive faults here in do_exit. Safest is to just
752  	 * leave this task alone and wait for reboot.
753  	 */
754  	if (unlikely(tsk->flags & PF_EXITING)) {
755  		pr_alert("Fixing recursive fault but reboot is needed!\n");
756  		futex_exit_recursive(tsk);
757  		set_current_state(TASK_UNINTERRUPTIBLE);
758  		schedule();
759  	}
760  
761  	exit_signals(tsk);  /* sets PF_EXITING */
762  
763  	/* sync mm's RSS info before statistics gathering */
764  	if (tsk->mm)
765  		sync_mm_rss(tsk->mm);
766  	acct_update_integrals(tsk);
767  	group_dead = atomic_dec_and_test(&tsk->signal->live);
768  	if (group_dead) {
769  		/*
770  		 * If the last thread of global init has exited, panic
771  		 * immediately to get a useable coredump.
772  		 */
773  		if (unlikely(is_global_init(tsk)))
774  			panic("Attempted to kill init! exitcode=0x%08x\n",
775  				tsk->signal->group_exit_code ?: (int)code);
776  
777  #ifdef CONFIG_POSIX_TIMERS
778  		hrtimer_cancel(&tsk->signal->real_timer);
779  		exit_itimers(tsk->signal);
780  #endif
781  		if (tsk->mm)
782  			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
783  	}
784  	acct_collect(code, group_dead);
785  	if (group_dead)
786  		tty_audit_exit();
787  	audit_free(tsk);
788  
789  	tsk->exit_code = code;
790  	taskstats_exit(tsk, group_dead);
791  
792  	exit_mm();
793  
794  	if (group_dead)
795  		acct_process();
796  	trace_sched_process_exit(tsk);
797  
798  	exit_sem(tsk);
799  	exit_shm(tsk);
800  	exit_files(tsk);
801  	exit_fs(tsk);
802  	if (group_dead)
803  		disassociate_ctty(1);
804  	exit_task_namespaces(tsk);
805  	exit_task_work(tsk);
806  	exit_thread(tsk);
807  	exit_umh(tsk);
808  
809  	/*
810  	 * Flush inherited counters to the parent - before the parent
811  	 * gets woken up by child-exit notifications.
812  	 *
813  	 * because of cgroup mode, must be called before cgroup_exit()
814  	 */
815  	perf_event_exit_task(tsk);
816  
817  	sched_autogroup_exit_task(tsk);
818  	cgroup_exit(tsk);
819  
820  	/*
821  	 * FIXME: do that only when needed, using sched_exit tracepoint
822  	 */
823  	flush_ptrace_hw_breakpoint(tsk);
824  
825  	exit_tasks_rcu_start();
826  	exit_notify(tsk, group_dead);
827  	proc_exit_connector(tsk);
828  	mpol_put_task_policy(tsk);
829  #ifdef CONFIG_FUTEX
830  	if (unlikely(current->pi_state_cache))
831  		kfree(current->pi_state_cache);
832  #endif
833  	/*
834  	 * Make sure we are holding no locks:
835  	 */
836  	debug_check_no_locks_held();
837  
838  	if (tsk->io_context)
839  		exit_io_context(tsk);
840  
841  	if (tsk->splice_pipe)
842  		free_pipe_info(tsk->splice_pipe);
843  
844  	if (tsk->task_frag.page)
845  		put_page(tsk->task_frag.page);
846  
847  	validate_creds_for_do_exit(tsk);
848  
849  	check_stack_usage();
850  	preempt_disable();
851  	if (tsk->nr_dirtied)
852  		__this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
853  	exit_rcu();
854  	exit_tasks_rcu_finish();
855  
856  	lockdep_free_task(tsk);
857  	do_task_dead();
858  }
859  EXPORT_SYMBOL_GPL(do_exit);
860  
861  void complete_and_exit(struct completion *comp, long code)
862  {
863  	if (comp)
864  		complete(comp);
865  
866  	do_exit(code);
867  }
868  EXPORT_SYMBOL(complete_and_exit);
869  
870  SYSCALL_DEFINE1(exit, int, error_code)
871  {
872  	do_exit((error_code&0xff)<<8);
873  }
874  
875  /*
876   * Take down every thread in the group.  This is called by fatal signals
877   * as well as by sys_exit_group (below).
878   */
879  void
880  do_group_exit(int exit_code)
881  {
882  	struct signal_struct *sig = current->signal;
883  
884  	BUG_ON(exit_code & 0x80); /* core dumps don't get here */
885  
886  	if (signal_group_exit(sig))
887  		exit_code = sig->group_exit_code;
888  	else if (!thread_group_empty(current)) {
889  		struct sighand_struct *const sighand = current->sighand;
890  
891  		spin_lock_irq(&sighand->siglock);
892  		if (signal_group_exit(sig))
893  			/* Another thread got here before we took the lock.  */
894  			exit_code = sig->group_exit_code;
895  		else {
896  			sig->group_exit_code = exit_code;
897  			sig->flags = SIGNAL_GROUP_EXIT;
898  			zap_other_threads(current);
899  		}
900  		spin_unlock_irq(&sighand->siglock);
901  	}
902  
903  	do_exit(exit_code);
904  	/* NOTREACHED */
905  }
906  
907  /*
908   * this kills every thread in the thread group. Note that any externally
909   * wait4()-ing process will get the correct exit code - even if this
910   * thread is not the thread group leader.
911   */
912  SYSCALL_DEFINE1(exit_group, int, error_code)
913  {
914  	do_group_exit((error_code & 0xff) << 8);
915  	/* NOTREACHED */
916  	return 0;
917  }
918  
919  struct waitid_info {
920  	pid_t pid;
921  	uid_t uid;
922  	int status;
923  	int cause;
924  };
925  
926  struct wait_opts {
927  	enum pid_type		wo_type;
928  	int			wo_flags;
929  	struct pid		*wo_pid;
930  
931  	struct waitid_info	*wo_info;
932  	int			wo_stat;
933  	struct rusage		*wo_rusage;
934  
935  	wait_queue_entry_t		child_wait;
936  	int			notask_error;
937  };
938  
939  static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
940  {
941  	return	wo->wo_type == PIDTYPE_MAX ||
942  		task_pid_type(p, wo->wo_type) == wo->wo_pid;
943  }
944  
945  static int
946  eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
947  {
948  	if (!eligible_pid(wo, p))
949  		return 0;
950  
951  	/*
952  	 * Wait for all children (clone and not) if __WALL is set or
953  	 * if it is traced by us.
954  	 */
955  	if (ptrace || (wo->wo_flags & __WALL))
956  		return 1;
957  
958  	/*
959  	 * Otherwise, wait for clone children *only* if __WCLONE is set;
960  	 * otherwise, wait for non-clone children *only*.
961  	 *
962  	 * Note: a "clone" child here is one that reports to its parent
963  	 * using a signal other than SIGCHLD, or a non-leader thread which
964  	 * we can only see if it is traced by us.
965  	 */
966  	if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
967  		return 0;
968  
969  	return 1;
970  }
971  
972  /*
973   * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
974   * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
975   * the lock and this task is uninteresting.  If we return nonzero, we have
976   * released the lock and the system call should return.
977   */
978  static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
979  {
980  	int state, status;
981  	pid_t pid = task_pid_vnr(p);
982  	uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
983  	struct waitid_info *infop;
984  
985  	if (!likely(wo->wo_flags & WEXITED))
986  		return 0;
987  
988  	if (unlikely(wo->wo_flags & WNOWAIT)) {
989  		status = p->exit_code;
990  		get_task_struct(p);
991  		read_unlock(&tasklist_lock);
992  		sched_annotate_sleep();
993  		if (wo->wo_rusage)
994  			getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
995  		put_task_struct(p);
996  		goto out_info;
997  	}
998  	/*
999  	 * Move the task's state to DEAD/TRACE, only one thread can do this.
1000  	 */
1001  	state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1002  		EXIT_TRACE : EXIT_DEAD;
1003  	if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1004  		return 0;
1005  	/*
1006  	 * We own this thread, nobody else can reap it.
1007  	 */
1008  	read_unlock(&tasklist_lock);
1009  	sched_annotate_sleep();
1010  
1011  	/*
1012  	 * Check thread_group_leader() to exclude the traced sub-threads.
1013  	 */
1014  	if (state == EXIT_DEAD && thread_group_leader(p)) {
1015  		struct signal_struct *sig = p->signal;
1016  		struct signal_struct *psig = current->signal;
1017  		unsigned long maxrss;
1018  		u64 tgutime, tgstime;
1019  
1020  		/*
1021  		 * The resource counters for the group leader are in its
1022  		 * own task_struct.  Those for dead threads in the group
1023  		 * are in its signal_struct, as are those for the child
1024  		 * processes it has previously reaped.  All these
1025  		 * accumulate in the parent's signal_struct c* fields.
1026  		 *
1027  		 * We don't bother to take a lock here to protect these
1028  		 * p->signal fields because the whole thread group is dead
1029  		 * and nobody can change them.
1030  		 *
1031  		 * psig->stats_lock also protects us from our sub-theads
1032  		 * which can reap other children at the same time. Until
1033  		 * we change k_getrusage()-like users to rely on this lock
1034  		 * we have to take ->siglock as well.
1035  		 *
1036  		 * We use thread_group_cputime_adjusted() to get times for
1037  		 * the thread group, which consolidates times for all threads
1038  		 * in the group including the group leader.
1039  		 */
1040  		thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1041  		spin_lock_irq(&current->sighand->siglock);
1042  		write_seqlock(&psig->stats_lock);
1043  		psig->cutime += tgutime + sig->cutime;
1044  		psig->cstime += tgstime + sig->cstime;
1045  		psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1046  		psig->cmin_flt +=
1047  			p->min_flt + sig->min_flt + sig->cmin_flt;
1048  		psig->cmaj_flt +=
1049  			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1050  		psig->cnvcsw +=
1051  			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1052  		psig->cnivcsw +=
1053  			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1054  		psig->cinblock +=
1055  			task_io_get_inblock(p) +
1056  			sig->inblock + sig->cinblock;
1057  		psig->coublock +=
1058  			task_io_get_oublock(p) +
1059  			sig->oublock + sig->coublock;
1060  		maxrss = max(sig->maxrss, sig->cmaxrss);
1061  		if (psig->cmaxrss < maxrss)
1062  			psig->cmaxrss = maxrss;
1063  		task_io_accounting_add(&psig->ioac, &p->ioac);
1064  		task_io_accounting_add(&psig->ioac, &sig->ioac);
1065  		write_sequnlock(&psig->stats_lock);
1066  		spin_unlock_irq(&current->sighand->siglock);
1067  	}
1068  
1069  	if (wo->wo_rusage)
1070  		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1071  	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1072  		? p->signal->group_exit_code : p->exit_code;
1073  	wo->wo_stat = status;
1074  
1075  	if (state == EXIT_TRACE) {
1076  		write_lock_irq(&tasklist_lock);
1077  		/* We dropped tasklist, ptracer could die and untrace */
1078  		ptrace_unlink(p);
1079  
1080  		/* If parent wants a zombie, don't release it now */
1081  		state = EXIT_ZOMBIE;
1082  		if (do_notify_parent(p, p->exit_signal))
1083  			state = EXIT_DEAD;
1084  		p->exit_state = state;
1085  		write_unlock_irq(&tasklist_lock);
1086  	}
1087  	if (state == EXIT_DEAD)
1088  		release_task(p);
1089  
1090  out_info:
1091  	infop = wo->wo_info;
1092  	if (infop) {
1093  		if ((status & 0x7f) == 0) {
1094  			infop->cause = CLD_EXITED;
1095  			infop->status = status >> 8;
1096  		} else {
1097  			infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1098  			infop->status = status & 0x7f;
1099  		}
1100  		infop->pid = pid;
1101  		infop->uid = uid;
1102  	}
1103  
1104  	return pid;
1105  }
1106  
1107  static int *task_stopped_code(struct task_struct *p, bool ptrace)
1108  {
1109  	if (ptrace) {
1110  		if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1111  			return &p->exit_code;
1112  	} else {
1113  		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1114  			return &p->signal->group_exit_code;
1115  	}
1116  	return NULL;
1117  }
1118  
1119  /**
1120   * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1121   * @wo: wait options
1122   * @ptrace: is the wait for ptrace
1123   * @p: task to wait for
1124   *
1125   * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1126   *
1127   * CONTEXT:
1128   * read_lock(&tasklist_lock), which is released if return value is
1129   * non-zero.  Also, grabs and releases @p->sighand->siglock.
1130   *
1131   * RETURNS:
1132   * 0 if wait condition didn't exist and search for other wait conditions
1133   * should continue.  Non-zero return, -errno on failure and @p's pid on
1134   * success, implies that tasklist_lock is released and wait condition
1135   * search should terminate.
1136   */
1137  static int wait_task_stopped(struct wait_opts *wo,
1138  				int ptrace, struct task_struct *p)
1139  {
1140  	struct waitid_info *infop;
1141  	int exit_code, *p_code, why;
1142  	uid_t uid = 0; /* unneeded, required by compiler */
1143  	pid_t pid;
1144  
1145  	/*
1146  	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1147  	 */
1148  	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1149  		return 0;
1150  
1151  	if (!task_stopped_code(p, ptrace))
1152  		return 0;
1153  
1154  	exit_code = 0;
1155  	spin_lock_irq(&p->sighand->siglock);
1156  
1157  	p_code = task_stopped_code(p, ptrace);
1158  	if (unlikely(!p_code))
1159  		goto unlock_sig;
1160  
1161  	exit_code = *p_code;
1162  	if (!exit_code)
1163  		goto unlock_sig;
1164  
1165  	if (!unlikely(wo->wo_flags & WNOWAIT))
1166  		*p_code = 0;
1167  
1168  	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1169  unlock_sig:
1170  	spin_unlock_irq(&p->sighand->siglock);
1171  	if (!exit_code)
1172  		return 0;
1173  
1174  	/*
1175  	 * Now we are pretty sure this task is interesting.
1176  	 * Make sure it doesn't get reaped out from under us while we
1177  	 * give up the lock and then examine it below.  We don't want to
1178  	 * keep holding onto the tasklist_lock while we call getrusage and
1179  	 * possibly take page faults for user memory.
1180  	 */
1181  	get_task_struct(p);
1182  	pid = task_pid_vnr(p);
1183  	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1184  	read_unlock(&tasklist_lock);
1185  	sched_annotate_sleep();
1186  	if (wo->wo_rusage)
1187  		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1188  	put_task_struct(p);
1189  
1190  	if (likely(!(wo->wo_flags & WNOWAIT)))
1191  		wo->wo_stat = (exit_code << 8) | 0x7f;
1192  
1193  	infop = wo->wo_info;
1194  	if (infop) {
1195  		infop->cause = why;
1196  		infop->status = exit_code;
1197  		infop->pid = pid;
1198  		infop->uid = uid;
1199  	}
1200  	return pid;
1201  }
1202  
1203  /*
1204   * Handle do_wait work for one task in a live, non-stopped state.
1205   * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1206   * the lock and this task is uninteresting.  If we return nonzero, we have
1207   * released the lock and the system call should return.
1208   */
1209  static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1210  {
1211  	struct waitid_info *infop;
1212  	pid_t pid;
1213  	uid_t uid;
1214  
1215  	if (!unlikely(wo->wo_flags & WCONTINUED))
1216  		return 0;
1217  
1218  	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1219  		return 0;
1220  
1221  	spin_lock_irq(&p->sighand->siglock);
1222  	/* Re-check with the lock held.  */
1223  	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1224  		spin_unlock_irq(&p->sighand->siglock);
1225  		return 0;
1226  	}
1227  	if (!unlikely(wo->wo_flags & WNOWAIT))
1228  		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1229  	uid = from_kuid_munged(current_user_ns(), task_uid(p));
1230  	spin_unlock_irq(&p->sighand->siglock);
1231  
1232  	pid = task_pid_vnr(p);
1233  	get_task_struct(p);
1234  	read_unlock(&tasklist_lock);
1235  	sched_annotate_sleep();
1236  	if (wo->wo_rusage)
1237  		getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1238  	put_task_struct(p);
1239  
1240  	infop = wo->wo_info;
1241  	if (!infop) {
1242  		wo->wo_stat = 0xffff;
1243  	} else {
1244  		infop->cause = CLD_CONTINUED;
1245  		infop->pid = pid;
1246  		infop->uid = uid;
1247  		infop->status = SIGCONT;
1248  	}
1249  	return pid;
1250  }
1251  
1252  /*
1253   * Consider @p for a wait by @parent.
1254   *
1255   * -ECHILD should be in ->notask_error before the first call.
1256   * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1257   * Returns zero if the search for a child should continue;
1258   * then ->notask_error is 0 if @p is an eligible child,
1259   * or still -ECHILD.
1260   */
1261  static int wait_consider_task(struct wait_opts *wo, int ptrace,
1262  				struct task_struct *p)
1263  {
1264  	/*
1265  	 * We can race with wait_task_zombie() from another thread.
1266  	 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1267  	 * can't confuse the checks below.
1268  	 */
1269  	int exit_state = READ_ONCE(p->exit_state);
1270  	int ret;
1271  
1272  	if (unlikely(exit_state == EXIT_DEAD))
1273  		return 0;
1274  
1275  	ret = eligible_child(wo, ptrace, p);
1276  	if (!ret)
1277  		return ret;
1278  
1279  	if (unlikely(exit_state == EXIT_TRACE)) {
1280  		/*
1281  		 * ptrace == 0 means we are the natural parent. In this case
1282  		 * we should clear notask_error, debugger will notify us.
1283  		 */
1284  		if (likely(!ptrace))
1285  			wo->notask_error = 0;
1286  		return 0;
1287  	}
1288  
1289  	if (likely(!ptrace) && unlikely(p->ptrace)) {
1290  		/*
1291  		 * If it is traced by its real parent's group, just pretend
1292  		 * the caller is ptrace_do_wait() and reap this child if it
1293  		 * is zombie.
1294  		 *
1295  		 * This also hides group stop state from real parent; otherwise
1296  		 * a single stop can be reported twice as group and ptrace stop.
1297  		 * If a ptracer wants to distinguish these two events for its
1298  		 * own children it should create a separate process which takes
1299  		 * the role of real parent.
1300  		 */
1301  		if (!ptrace_reparented(p))
1302  			ptrace = 1;
1303  	}
1304  
1305  	/* slay zombie? */
1306  	if (exit_state == EXIT_ZOMBIE) {
1307  		/* we don't reap group leaders with subthreads */
1308  		if (!delay_group_leader(p)) {
1309  			/*
1310  			 * A zombie ptracee is only visible to its ptracer.
1311  			 * Notification and reaping will be cascaded to the
1312  			 * real parent when the ptracer detaches.
1313  			 */
1314  			if (unlikely(ptrace) || likely(!p->ptrace))
1315  				return wait_task_zombie(wo, p);
1316  		}
1317  
1318  		/*
1319  		 * Allow access to stopped/continued state via zombie by
1320  		 * falling through.  Clearing of notask_error is complex.
1321  		 *
1322  		 * When !@ptrace:
1323  		 *
1324  		 * If WEXITED is set, notask_error should naturally be
1325  		 * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1326  		 * so, if there are live subthreads, there are events to
1327  		 * wait for.  If all subthreads are dead, it's still safe
1328  		 * to clear - this function will be called again in finite
1329  		 * amount time once all the subthreads are released and
1330  		 * will then return without clearing.
1331  		 *
1332  		 * When @ptrace:
1333  		 *
1334  		 * Stopped state is per-task and thus can't change once the
1335  		 * target task dies.  Only continued and exited can happen.
1336  		 * Clear notask_error if WCONTINUED | WEXITED.
1337  		 */
1338  		if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1339  			wo->notask_error = 0;
1340  	} else {
1341  		/*
1342  		 * @p is alive and it's gonna stop, continue or exit, so
1343  		 * there always is something to wait for.
1344  		 */
1345  		wo->notask_error = 0;
1346  	}
1347  
1348  	/*
1349  	 * Wait for stopped.  Depending on @ptrace, different stopped state
1350  	 * is used and the two don't interact with each other.
1351  	 */
1352  	ret = wait_task_stopped(wo, ptrace, p);
1353  	if (ret)
1354  		return ret;
1355  
1356  	/*
1357  	 * Wait for continued.  There's only one continued state and the
1358  	 * ptracer can consume it which can confuse the real parent.  Don't
1359  	 * use WCONTINUED from ptracer.  You don't need or want it.
1360  	 */
1361  	return wait_task_continued(wo, p);
1362  }
1363  
1364  /*
1365   * Do the work of do_wait() for one thread in the group, @tsk.
1366   *
1367   * -ECHILD should be in ->notask_error before the first call.
1368   * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1369   * Returns zero if the search for a child should continue; then
1370   * ->notask_error is 0 if there were any eligible children,
1371   * or still -ECHILD.
1372   */
1373  static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1374  {
1375  	struct task_struct *p;
1376  
1377  	list_for_each_entry(p, &tsk->children, sibling) {
1378  		int ret = wait_consider_task(wo, 0, p);
1379  
1380  		if (ret)
1381  			return ret;
1382  	}
1383  
1384  	return 0;
1385  }
1386  
1387  static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1388  {
1389  	struct task_struct *p;
1390  
1391  	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1392  		int ret = wait_consider_task(wo, 1, p);
1393  
1394  		if (ret)
1395  			return ret;
1396  	}
1397  
1398  	return 0;
1399  }
1400  
1401  static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1402  				int sync, void *key)
1403  {
1404  	struct wait_opts *wo = container_of(wait, struct wait_opts,
1405  						child_wait);
1406  	struct task_struct *p = key;
1407  
1408  	if (!eligible_pid(wo, p))
1409  		return 0;
1410  
1411  	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1412  		return 0;
1413  
1414  	return default_wake_function(wait, mode, sync, key);
1415  }
1416  
1417  void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1418  {
1419  	__wake_up_sync_key(&parent->signal->wait_chldexit,
1420  			   TASK_INTERRUPTIBLE, p);
1421  }
1422  
1423  static long do_wait(struct wait_opts *wo)
1424  {
1425  	struct task_struct *tsk;
1426  	int retval;
1427  
1428  	trace_sched_process_wait(wo->wo_pid);
1429  
1430  	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1431  	wo->child_wait.private = current;
1432  	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1433  repeat:
1434  	/*
1435  	 * If there is nothing that can match our criteria, just get out.
1436  	 * We will clear ->notask_error to zero if we see any child that
1437  	 * might later match our criteria, even if we are not able to reap
1438  	 * it yet.
1439  	 */
1440  	wo->notask_error = -ECHILD;
1441  	if ((wo->wo_type < PIDTYPE_MAX) &&
1442  	   (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1443  		goto notask;
1444  
1445  	set_current_state(TASK_INTERRUPTIBLE);
1446  	read_lock(&tasklist_lock);
1447  	tsk = current;
1448  	do {
1449  		retval = do_wait_thread(wo, tsk);
1450  		if (retval)
1451  			goto end;
1452  
1453  		retval = ptrace_do_wait(wo, tsk);
1454  		if (retval)
1455  			goto end;
1456  
1457  		if (wo->wo_flags & __WNOTHREAD)
1458  			break;
1459  	} while_each_thread(current, tsk);
1460  	read_unlock(&tasklist_lock);
1461  
1462  notask:
1463  	retval = wo->notask_error;
1464  	if (!retval && !(wo->wo_flags & WNOHANG)) {
1465  		retval = -ERESTARTSYS;
1466  		if (!signal_pending(current)) {
1467  			schedule();
1468  			goto repeat;
1469  		}
1470  	}
1471  end:
1472  	__set_current_state(TASK_RUNNING);
1473  	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1474  	return retval;
1475  }
1476  
1477  static struct pid *pidfd_get_pid(unsigned int fd)
1478  {
1479  	struct fd f;
1480  	struct pid *pid;
1481  
1482  	f = fdget(fd);
1483  	if (!f.file)
1484  		return ERR_PTR(-EBADF);
1485  
1486  	pid = pidfd_pid(f.file);
1487  	if (!IS_ERR(pid))
1488  		get_pid(pid);
1489  
1490  	fdput(f);
1491  	return pid;
1492  }
1493  
1494  static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1495  			  int options, struct rusage *ru)
1496  {
1497  	struct wait_opts wo;
1498  	struct pid *pid = NULL;
1499  	enum pid_type type;
1500  	long ret;
1501  
1502  	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1503  			__WNOTHREAD|__WCLONE|__WALL))
1504  		return -EINVAL;
1505  	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1506  		return -EINVAL;
1507  
1508  	switch (which) {
1509  	case P_ALL:
1510  		type = PIDTYPE_MAX;
1511  		break;
1512  	case P_PID:
1513  		type = PIDTYPE_PID;
1514  		if (upid <= 0)
1515  			return -EINVAL;
1516  
1517  		pid = find_get_pid(upid);
1518  		break;
1519  	case P_PGID:
1520  		type = PIDTYPE_PGID;
1521  		if (upid < 0)
1522  			return -EINVAL;
1523  
1524  		if (upid)
1525  			pid = find_get_pid(upid);
1526  		else
1527  			pid = get_task_pid(current, PIDTYPE_PGID);
1528  		break;
1529  	case P_PIDFD:
1530  		type = PIDTYPE_PID;
1531  		if (upid < 0)
1532  			return -EINVAL;
1533  
1534  		pid = pidfd_get_pid(upid);
1535  		if (IS_ERR(pid))
1536  			return PTR_ERR(pid);
1537  		break;
1538  	default:
1539  		return -EINVAL;
1540  	}
1541  
1542  	wo.wo_type	= type;
1543  	wo.wo_pid	= pid;
1544  	wo.wo_flags	= options;
1545  	wo.wo_info	= infop;
1546  	wo.wo_rusage	= ru;
1547  	ret = do_wait(&wo);
1548  
1549  	put_pid(pid);
1550  	return ret;
1551  }
1552  
1553  SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1554  		infop, int, options, struct rusage __user *, ru)
1555  {
1556  	struct rusage r;
1557  	struct waitid_info info = {.status = 0};
1558  	long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1559  	int signo = 0;
1560  
1561  	if (err > 0) {
1562  		signo = SIGCHLD;
1563  		err = 0;
1564  		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1565  			return -EFAULT;
1566  	}
1567  	if (!infop)
1568  		return err;
1569  
1570  	if (!user_write_access_begin(infop, sizeof(*infop)))
1571  		return -EFAULT;
1572  
1573  	unsafe_put_user(signo, &infop->si_signo, Efault);
1574  	unsafe_put_user(0, &infop->si_errno, Efault);
1575  	unsafe_put_user(info.cause, &infop->si_code, Efault);
1576  	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1577  	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1578  	unsafe_put_user(info.status, &infop->si_status, Efault);
1579  	user_write_access_end();
1580  	return err;
1581  Efault:
1582  	user_write_access_end();
1583  	return -EFAULT;
1584  }
1585  
1586  long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1587  		  struct rusage *ru)
1588  {
1589  	struct wait_opts wo;
1590  	struct pid *pid = NULL;
1591  	enum pid_type type;
1592  	long ret;
1593  
1594  	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1595  			__WNOTHREAD|__WCLONE|__WALL))
1596  		return -EINVAL;
1597  
1598  	/* -INT_MIN is not defined */
1599  	if (upid == INT_MIN)
1600  		return -ESRCH;
1601  
1602  	if (upid == -1)
1603  		type = PIDTYPE_MAX;
1604  	else if (upid < 0) {
1605  		type = PIDTYPE_PGID;
1606  		pid = find_get_pid(-upid);
1607  	} else if (upid == 0) {
1608  		type = PIDTYPE_PGID;
1609  		pid = get_task_pid(current, PIDTYPE_PGID);
1610  	} else /* upid > 0 */ {
1611  		type = PIDTYPE_PID;
1612  		pid = find_get_pid(upid);
1613  	}
1614  
1615  	wo.wo_type	= type;
1616  	wo.wo_pid	= pid;
1617  	wo.wo_flags	= options | WEXITED;
1618  	wo.wo_info	= NULL;
1619  	wo.wo_stat	= 0;
1620  	wo.wo_rusage	= ru;
1621  	ret = do_wait(&wo);
1622  	put_pid(pid);
1623  	if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1624  		ret = -EFAULT;
1625  
1626  	return ret;
1627  }
1628  
1629  SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1630  		int, options, struct rusage __user *, ru)
1631  {
1632  	struct rusage r;
1633  	long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1634  
1635  	if (err > 0) {
1636  		if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1637  			return -EFAULT;
1638  	}
1639  	return err;
1640  }
1641  
1642  #ifdef __ARCH_WANT_SYS_WAITPID
1643  
1644  /*
1645   * sys_waitpid() remains for compatibility. waitpid() should be
1646   * implemented by calling sys_wait4() from libc.a.
1647   */
1648  SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1649  {
1650  	return kernel_wait4(pid, stat_addr, options, NULL);
1651  }
1652  
1653  #endif
1654  
1655  #ifdef CONFIG_COMPAT
1656  COMPAT_SYSCALL_DEFINE4(wait4,
1657  	compat_pid_t, pid,
1658  	compat_uint_t __user *, stat_addr,
1659  	int, options,
1660  	struct compat_rusage __user *, ru)
1661  {
1662  	struct rusage r;
1663  	long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1664  	if (err > 0) {
1665  		if (ru && put_compat_rusage(&r, ru))
1666  			return -EFAULT;
1667  	}
1668  	return err;
1669  }
1670  
1671  COMPAT_SYSCALL_DEFINE5(waitid,
1672  		int, which, compat_pid_t, pid,
1673  		struct compat_siginfo __user *, infop, int, options,
1674  		struct compat_rusage __user *, uru)
1675  {
1676  	struct rusage ru;
1677  	struct waitid_info info = {.status = 0};
1678  	long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1679  	int signo = 0;
1680  	if (err > 0) {
1681  		signo = SIGCHLD;
1682  		err = 0;
1683  		if (uru) {
1684  			/* kernel_waitid() overwrites everything in ru */
1685  			if (COMPAT_USE_64BIT_TIME)
1686  				err = copy_to_user(uru, &ru, sizeof(ru));
1687  			else
1688  				err = put_compat_rusage(&ru, uru);
1689  			if (err)
1690  				return -EFAULT;
1691  		}
1692  	}
1693  
1694  	if (!infop)
1695  		return err;
1696  
1697  	if (!user_write_access_begin(infop, sizeof(*infop)))
1698  		return -EFAULT;
1699  
1700  	unsafe_put_user(signo, &infop->si_signo, Efault);
1701  	unsafe_put_user(0, &infop->si_errno, Efault);
1702  	unsafe_put_user(info.cause, &infop->si_code, Efault);
1703  	unsafe_put_user(info.pid, &infop->si_pid, Efault);
1704  	unsafe_put_user(info.uid, &infop->si_uid, Efault);
1705  	unsafe_put_user(info.status, &infop->si_status, Efault);
1706  	user_write_access_end();
1707  	return err;
1708  Efault:
1709  	user_write_access_end();
1710  	return -EFAULT;
1711  }
1712  #endif
1713  
1714  __weak void abort(void)
1715  {
1716  	BUG();
1717  
1718  	/* if that doesn't kill us, halt */
1719  	panic("Oops failed to kill thread");
1720  }
1721  EXPORT_SYMBOL(abort);
1722