xref: /openbmc/linux/kernel/exit.c (revision b6dcefde)
1 /*
2  *  linux/kernel/exit.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/mm.h>
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/iocontext.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/fdtable.h>
23 #include <linux/binfmts.h>
24 #include <linux/nsproxy.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/ptrace.h>
27 #include <linux/profile.h>
28 #include <linux/mount.h>
29 #include <linux/proc_fs.h>
30 #include <linux/kthread.h>
31 #include <linux/mempolicy.h>
32 #include <linux/taskstats_kern.h>
33 #include <linux/delayacct.h>
34 #include <linux/freezer.h>
35 #include <linux/cgroup.h>
36 #include <linux/syscalls.h>
37 #include <linux/signal.h>
38 #include <linux/posix-timers.h>
39 #include <linux/cn_proc.h>
40 #include <linux/mutex.h>
41 #include <linux/futex.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/tracehook.h>
48 #include <linux/fs_struct.h>
49 #include <linux/init_task.h>
50 #include <linux/perf_event.h>
51 #include <trace/events/sched.h>
52 #include <linux/hw_breakpoint.h>
53 
54 #include <asm/uaccess.h>
55 #include <asm/unistd.h>
56 #include <asm/pgtable.h>
57 #include <asm/mmu_context.h>
58 #include "cred-internals.h"
59 
60 static void exit_mm(struct task_struct * tsk);
61 
62 static void __unhash_process(struct task_struct *p)
63 {
64 	nr_threads--;
65 	detach_pid(p, PIDTYPE_PID);
66 	if (thread_group_leader(p)) {
67 		detach_pid(p, PIDTYPE_PGID);
68 		detach_pid(p, PIDTYPE_SID);
69 
70 		list_del_rcu(&p->tasks);
71 		list_del_init(&p->sibling);
72 		__get_cpu_var(process_counts)--;
73 	}
74 	list_del_rcu(&p->thread_group);
75 }
76 
77 /*
78  * This function expects the tasklist_lock write-locked.
79  */
80 static void __exit_signal(struct task_struct *tsk)
81 {
82 	struct signal_struct *sig = tsk->signal;
83 	struct sighand_struct *sighand;
84 
85 	BUG_ON(!sig);
86 	BUG_ON(!atomic_read(&sig->count));
87 
88 	sighand = rcu_dereference(tsk->sighand);
89 	spin_lock(&sighand->siglock);
90 
91 	posix_cpu_timers_exit(tsk);
92 	if (atomic_dec_and_test(&sig->count))
93 		posix_cpu_timers_exit_group(tsk);
94 	else {
95 		/*
96 		 * If there is any task waiting for the group exit
97 		 * then notify it:
98 		 */
99 		if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
100 			wake_up_process(sig->group_exit_task);
101 
102 		if (tsk == sig->curr_target)
103 			sig->curr_target = next_thread(tsk);
104 		/*
105 		 * Accumulate here the counters for all threads but the
106 		 * group leader as they die, so they can be added into
107 		 * the process-wide totals when those are taken.
108 		 * The group leader stays around as a zombie as long
109 		 * as there are other threads.  When it gets reaped,
110 		 * the exit.c code will add its counts into these totals.
111 		 * We won't ever get here for the group leader, since it
112 		 * will have been the last reference on the signal_struct.
113 		 */
114 		sig->utime = cputime_add(sig->utime, tsk->utime);
115 		sig->stime = cputime_add(sig->stime, tsk->stime);
116 		sig->gtime = cputime_add(sig->gtime, tsk->gtime);
117 		sig->min_flt += tsk->min_flt;
118 		sig->maj_flt += tsk->maj_flt;
119 		sig->nvcsw += tsk->nvcsw;
120 		sig->nivcsw += tsk->nivcsw;
121 		sig->inblock += task_io_get_inblock(tsk);
122 		sig->oublock += task_io_get_oublock(tsk);
123 		task_io_accounting_add(&sig->ioac, &tsk->ioac);
124 		sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
125 		sig = NULL; /* Marker for below. */
126 	}
127 
128 	__unhash_process(tsk);
129 
130 	/*
131 	 * Do this under ->siglock, we can race with another thread
132 	 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
133 	 */
134 	flush_sigqueue(&tsk->pending);
135 
136 	tsk->signal = NULL;
137 	tsk->sighand = NULL;
138 	spin_unlock(&sighand->siglock);
139 
140 	__cleanup_sighand(sighand);
141 	clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
142 	if (sig) {
143 		flush_sigqueue(&sig->shared_pending);
144 		taskstats_tgid_free(sig);
145 		/*
146 		 * Make sure ->signal can't go away under rq->lock,
147 		 * see account_group_exec_runtime().
148 		 */
149 		task_rq_unlock_wait(tsk);
150 		__cleanup_signal(sig);
151 	}
152 }
153 
154 static void delayed_put_task_struct(struct rcu_head *rhp)
155 {
156 	struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
157 
158 #ifdef CONFIG_PERF_EVENTS
159 	WARN_ON_ONCE(tsk->perf_event_ctxp);
160 #endif
161 	trace_sched_process_free(tsk);
162 	put_task_struct(tsk);
163 }
164 
165 
166 void release_task(struct task_struct * p)
167 {
168 	struct task_struct *leader;
169 	int zap_leader;
170 repeat:
171 	tracehook_prepare_release_task(p);
172 	/* don't need to get the RCU readlock here - the process is dead and
173 	 * can't be modifying its own credentials */
174 	atomic_dec(&__task_cred(p)->user->processes);
175 
176 	proc_flush_task(p);
177 
178 	write_lock_irq(&tasklist_lock);
179 	tracehook_finish_release_task(p);
180 	__exit_signal(p);
181 
182 	/*
183 	 * If we are the last non-leader member of the thread
184 	 * group, and the leader is zombie, then notify the
185 	 * group leader's parent process. (if it wants notification.)
186 	 */
187 	zap_leader = 0;
188 	leader = p->group_leader;
189 	if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
190 		BUG_ON(task_detached(leader));
191 		do_notify_parent(leader, leader->exit_signal);
192 		/*
193 		 * If we were the last child thread and the leader has
194 		 * exited already, and the leader's parent ignores SIGCHLD,
195 		 * then we are the one who should release the leader.
196 		 *
197 		 * do_notify_parent() will have marked it self-reaping in
198 		 * that case.
199 		 */
200 		zap_leader = task_detached(leader);
201 
202 		/*
203 		 * This maintains the invariant that release_task()
204 		 * only runs on a task in EXIT_DEAD, just for sanity.
205 		 */
206 		if (zap_leader)
207 			leader->exit_state = EXIT_DEAD;
208 	}
209 
210 	write_unlock_irq(&tasklist_lock);
211 	release_thread(p);
212 	call_rcu(&p->rcu, delayed_put_task_struct);
213 
214 	p = leader;
215 	if (unlikely(zap_leader))
216 		goto repeat;
217 }
218 
219 /*
220  * This checks not only the pgrp, but falls back on the pid if no
221  * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
222  * without this...
223  *
224  * The caller must hold rcu lock or the tasklist lock.
225  */
226 struct pid *session_of_pgrp(struct pid *pgrp)
227 {
228 	struct task_struct *p;
229 	struct pid *sid = NULL;
230 
231 	p = pid_task(pgrp, PIDTYPE_PGID);
232 	if (p == NULL)
233 		p = pid_task(pgrp, PIDTYPE_PID);
234 	if (p != NULL)
235 		sid = task_session(p);
236 
237 	return sid;
238 }
239 
240 /*
241  * Determine if a process group is "orphaned", according to the POSIX
242  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
243  * by terminal-generated stop signals.  Newly orphaned process groups are
244  * to receive a SIGHUP and a SIGCONT.
245  *
246  * "I ask you, have you ever known what it is to be an orphan?"
247  */
248 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
249 {
250 	struct task_struct *p;
251 
252 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
253 		if ((p == ignored_task) ||
254 		    (p->exit_state && thread_group_empty(p)) ||
255 		    is_global_init(p->real_parent))
256 			continue;
257 
258 		if (task_pgrp(p->real_parent) != pgrp &&
259 		    task_session(p->real_parent) == task_session(p))
260 			return 0;
261 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
262 
263 	return 1;
264 }
265 
266 int is_current_pgrp_orphaned(void)
267 {
268 	int retval;
269 
270 	read_lock(&tasklist_lock);
271 	retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
272 	read_unlock(&tasklist_lock);
273 
274 	return retval;
275 }
276 
277 static int has_stopped_jobs(struct pid *pgrp)
278 {
279 	int retval = 0;
280 	struct task_struct *p;
281 
282 	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
283 		if (!task_is_stopped(p))
284 			continue;
285 		retval = 1;
286 		break;
287 	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
288 	return retval;
289 }
290 
291 /*
292  * Check to see if any process groups have become orphaned as
293  * a result of our exiting, and if they have any stopped jobs,
294  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
295  */
296 static void
297 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
298 {
299 	struct pid *pgrp = task_pgrp(tsk);
300 	struct task_struct *ignored_task = tsk;
301 
302 	if (!parent)
303 		 /* exit: our father is in a different pgrp than
304 		  * we are and we were the only connection outside.
305 		  */
306 		parent = tsk->real_parent;
307 	else
308 		/* reparent: our child is in a different pgrp than
309 		 * we are, and it was the only connection outside.
310 		 */
311 		ignored_task = NULL;
312 
313 	if (task_pgrp(parent) != pgrp &&
314 	    task_session(parent) == task_session(tsk) &&
315 	    will_become_orphaned_pgrp(pgrp, ignored_task) &&
316 	    has_stopped_jobs(pgrp)) {
317 		__kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
318 		__kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
319 	}
320 }
321 
322 /**
323  * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
324  *
325  * If a kernel thread is launched as a result of a system call, or if
326  * it ever exits, it should generally reparent itself to kthreadd so it
327  * isn't in the way of other processes and is correctly cleaned up on exit.
328  *
329  * The various task state such as scheduling policy and priority may have
330  * been inherited from a user process, so we reset them to sane values here.
331  *
332  * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
333  */
334 static void reparent_to_kthreadd(void)
335 {
336 	write_lock_irq(&tasklist_lock);
337 
338 	ptrace_unlink(current);
339 	/* Reparent to init */
340 	current->real_parent = current->parent = kthreadd_task;
341 	list_move_tail(&current->sibling, &current->real_parent->children);
342 
343 	/* Set the exit signal to SIGCHLD so we signal init on exit */
344 	current->exit_signal = SIGCHLD;
345 
346 	if (task_nice(current) < 0)
347 		set_user_nice(current, 0);
348 	/* cpus_allowed? */
349 	/* rt_priority? */
350 	/* signals? */
351 	memcpy(current->signal->rlim, init_task.signal->rlim,
352 	       sizeof(current->signal->rlim));
353 
354 	atomic_inc(&init_cred.usage);
355 	commit_creds(&init_cred);
356 	write_unlock_irq(&tasklist_lock);
357 }
358 
359 void __set_special_pids(struct pid *pid)
360 {
361 	struct task_struct *curr = current->group_leader;
362 
363 	if (task_session(curr) != pid)
364 		change_pid(curr, PIDTYPE_SID, pid);
365 
366 	if (task_pgrp(curr) != pid)
367 		change_pid(curr, PIDTYPE_PGID, pid);
368 }
369 
370 static void set_special_pids(struct pid *pid)
371 {
372 	write_lock_irq(&tasklist_lock);
373 	__set_special_pids(pid);
374 	write_unlock_irq(&tasklist_lock);
375 }
376 
377 /*
378  * Let kernel threads use this to say that they allow a certain signal.
379  * Must not be used if kthread was cloned with CLONE_SIGHAND.
380  */
381 int allow_signal(int sig)
382 {
383 	if (!valid_signal(sig) || sig < 1)
384 		return -EINVAL;
385 
386 	spin_lock_irq(&current->sighand->siglock);
387 	/* This is only needed for daemonize()'ed kthreads */
388 	sigdelset(&current->blocked, sig);
389 	/*
390 	 * Kernel threads handle their own signals. Let the signal code
391 	 * know it'll be handled, so that they don't get converted to
392 	 * SIGKILL or just silently dropped.
393 	 */
394 	current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
395 	recalc_sigpending();
396 	spin_unlock_irq(&current->sighand->siglock);
397 	return 0;
398 }
399 
400 EXPORT_SYMBOL(allow_signal);
401 
402 int disallow_signal(int sig)
403 {
404 	if (!valid_signal(sig) || sig < 1)
405 		return -EINVAL;
406 
407 	spin_lock_irq(&current->sighand->siglock);
408 	current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
409 	recalc_sigpending();
410 	spin_unlock_irq(&current->sighand->siglock);
411 	return 0;
412 }
413 
414 EXPORT_SYMBOL(disallow_signal);
415 
416 /*
417  *	Put all the gunge required to become a kernel thread without
418  *	attached user resources in one place where it belongs.
419  */
420 
421 void daemonize(const char *name, ...)
422 {
423 	va_list args;
424 	sigset_t blocked;
425 
426 	va_start(args, name);
427 	vsnprintf(current->comm, sizeof(current->comm), name, args);
428 	va_end(args);
429 
430 	/*
431 	 * If we were started as result of loading a module, close all of the
432 	 * user space pages.  We don't need them, and if we didn't close them
433 	 * they would be locked into memory.
434 	 */
435 	exit_mm(current);
436 	/*
437 	 * We don't want to have TIF_FREEZE set if the system-wide hibernation
438 	 * or suspend transition begins right now.
439 	 */
440 	current->flags |= (PF_NOFREEZE | PF_KTHREAD);
441 
442 	if (current->nsproxy != &init_nsproxy) {
443 		get_nsproxy(&init_nsproxy);
444 		switch_task_namespaces(current, &init_nsproxy);
445 	}
446 	set_special_pids(&init_struct_pid);
447 	proc_clear_tty(current);
448 
449 	/* Block and flush all signals */
450 	sigfillset(&blocked);
451 	sigprocmask(SIG_BLOCK, &blocked, NULL);
452 	flush_signals(current);
453 
454 	/* Become as one with the init task */
455 
456 	daemonize_fs_struct();
457 	exit_files(current);
458 	current->files = init_task.files;
459 	atomic_inc(&current->files->count);
460 
461 	reparent_to_kthreadd();
462 }
463 
464 EXPORT_SYMBOL(daemonize);
465 
466 static void close_files(struct files_struct * files)
467 {
468 	int i, j;
469 	struct fdtable *fdt;
470 
471 	j = 0;
472 
473 	/*
474 	 * It is safe to dereference the fd table without RCU or
475 	 * ->file_lock because this is the last reference to the
476 	 * files structure.
477 	 */
478 	fdt = files_fdtable(files);
479 	for (;;) {
480 		unsigned long set;
481 		i = j * __NFDBITS;
482 		if (i >= fdt->max_fds)
483 			break;
484 		set = fdt->open_fds->fds_bits[j++];
485 		while (set) {
486 			if (set & 1) {
487 				struct file * file = xchg(&fdt->fd[i], NULL);
488 				if (file) {
489 					filp_close(file, files);
490 					cond_resched();
491 				}
492 			}
493 			i++;
494 			set >>= 1;
495 		}
496 	}
497 }
498 
499 struct files_struct *get_files_struct(struct task_struct *task)
500 {
501 	struct files_struct *files;
502 
503 	task_lock(task);
504 	files = task->files;
505 	if (files)
506 		atomic_inc(&files->count);
507 	task_unlock(task);
508 
509 	return files;
510 }
511 
512 void put_files_struct(struct files_struct *files)
513 {
514 	struct fdtable *fdt;
515 
516 	if (atomic_dec_and_test(&files->count)) {
517 		close_files(files);
518 		/*
519 		 * Free the fd and fdset arrays if we expanded them.
520 		 * If the fdtable was embedded, pass files for freeing
521 		 * at the end of the RCU grace period. Otherwise,
522 		 * you can free files immediately.
523 		 */
524 		fdt = files_fdtable(files);
525 		if (fdt != &files->fdtab)
526 			kmem_cache_free(files_cachep, files);
527 		free_fdtable(fdt);
528 	}
529 }
530 
531 void reset_files_struct(struct files_struct *files)
532 {
533 	struct task_struct *tsk = current;
534 	struct files_struct *old;
535 
536 	old = tsk->files;
537 	task_lock(tsk);
538 	tsk->files = files;
539 	task_unlock(tsk);
540 	put_files_struct(old);
541 }
542 
543 void exit_files(struct task_struct *tsk)
544 {
545 	struct files_struct * files = tsk->files;
546 
547 	if (files) {
548 		task_lock(tsk);
549 		tsk->files = NULL;
550 		task_unlock(tsk);
551 		put_files_struct(files);
552 	}
553 }
554 
555 #ifdef CONFIG_MM_OWNER
556 /*
557  * Task p is exiting and it owned mm, lets find a new owner for it
558  */
559 static inline int
560 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
561 {
562 	/*
563 	 * If there are other users of the mm and the owner (us) is exiting
564 	 * we need to find a new owner to take on the responsibility.
565 	 */
566 	if (atomic_read(&mm->mm_users) <= 1)
567 		return 0;
568 	if (mm->owner != p)
569 		return 0;
570 	return 1;
571 }
572 
573 void mm_update_next_owner(struct mm_struct *mm)
574 {
575 	struct task_struct *c, *g, *p = current;
576 
577 retry:
578 	if (!mm_need_new_owner(mm, p))
579 		return;
580 
581 	read_lock(&tasklist_lock);
582 	/*
583 	 * Search in the children
584 	 */
585 	list_for_each_entry(c, &p->children, sibling) {
586 		if (c->mm == mm)
587 			goto assign_new_owner;
588 	}
589 
590 	/*
591 	 * Search in the siblings
592 	 */
593 	list_for_each_entry(c, &p->real_parent->children, sibling) {
594 		if (c->mm == mm)
595 			goto assign_new_owner;
596 	}
597 
598 	/*
599 	 * Search through everything else. We should not get
600 	 * here often
601 	 */
602 	do_each_thread(g, c) {
603 		if (c->mm == mm)
604 			goto assign_new_owner;
605 	} while_each_thread(g, c);
606 
607 	read_unlock(&tasklist_lock);
608 	/*
609 	 * We found no owner yet mm_users > 1: this implies that we are
610 	 * most likely racing with swapoff (try_to_unuse()) or /proc or
611 	 * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
612 	 */
613 	mm->owner = NULL;
614 	return;
615 
616 assign_new_owner:
617 	BUG_ON(c == p);
618 	get_task_struct(c);
619 	/*
620 	 * The task_lock protects c->mm from changing.
621 	 * We always want mm->owner->mm == mm
622 	 */
623 	task_lock(c);
624 	/*
625 	 * Delay read_unlock() till we have the task_lock()
626 	 * to ensure that c does not slip away underneath us
627 	 */
628 	read_unlock(&tasklist_lock);
629 	if (c->mm != mm) {
630 		task_unlock(c);
631 		put_task_struct(c);
632 		goto retry;
633 	}
634 	mm->owner = c;
635 	task_unlock(c);
636 	put_task_struct(c);
637 }
638 #endif /* CONFIG_MM_OWNER */
639 
640 /*
641  * Turn us into a lazy TLB process if we
642  * aren't already..
643  */
644 static void exit_mm(struct task_struct * tsk)
645 {
646 	struct mm_struct *mm = tsk->mm;
647 	struct core_state *core_state;
648 
649 	mm_release(tsk, mm);
650 	if (!mm)
651 		return;
652 	/*
653 	 * Serialize with any possible pending coredump.
654 	 * We must hold mmap_sem around checking core_state
655 	 * and clearing tsk->mm.  The core-inducing thread
656 	 * will increment ->nr_threads for each thread in the
657 	 * group with ->mm != NULL.
658 	 */
659 	down_read(&mm->mmap_sem);
660 	core_state = mm->core_state;
661 	if (core_state) {
662 		struct core_thread self;
663 		up_read(&mm->mmap_sem);
664 
665 		self.task = tsk;
666 		self.next = xchg(&core_state->dumper.next, &self);
667 		/*
668 		 * Implies mb(), the result of xchg() must be visible
669 		 * to core_state->dumper.
670 		 */
671 		if (atomic_dec_and_test(&core_state->nr_threads))
672 			complete(&core_state->startup);
673 
674 		for (;;) {
675 			set_task_state(tsk, TASK_UNINTERRUPTIBLE);
676 			if (!self.task) /* see coredump_finish() */
677 				break;
678 			schedule();
679 		}
680 		__set_task_state(tsk, TASK_RUNNING);
681 		down_read(&mm->mmap_sem);
682 	}
683 	atomic_inc(&mm->mm_count);
684 	BUG_ON(mm != tsk->active_mm);
685 	/* more a memory barrier than a real lock */
686 	task_lock(tsk);
687 	tsk->mm = NULL;
688 	up_read(&mm->mmap_sem);
689 	enter_lazy_tlb(mm, current);
690 	/* We don't want this task to be frozen prematurely */
691 	clear_freeze_flag(tsk);
692 	task_unlock(tsk);
693 	mm_update_next_owner(mm);
694 	mmput(mm);
695 }
696 
697 /*
698  * When we die, we re-parent all our children.
699  * Try to give them to another thread in our thread
700  * group, and if no such member exists, give it to
701  * the child reaper process (ie "init") in our pid
702  * space.
703  */
704 static struct task_struct *find_new_reaper(struct task_struct *father)
705 {
706 	struct pid_namespace *pid_ns = task_active_pid_ns(father);
707 	struct task_struct *thread;
708 
709 	thread = father;
710 	while_each_thread(father, thread) {
711 		if (thread->flags & PF_EXITING)
712 			continue;
713 		if (unlikely(pid_ns->child_reaper == father))
714 			pid_ns->child_reaper = thread;
715 		return thread;
716 	}
717 
718 	if (unlikely(pid_ns->child_reaper == father)) {
719 		write_unlock_irq(&tasklist_lock);
720 		if (unlikely(pid_ns == &init_pid_ns))
721 			panic("Attempted to kill init!");
722 
723 		zap_pid_ns_processes(pid_ns);
724 		write_lock_irq(&tasklist_lock);
725 		/*
726 		 * We can not clear ->child_reaper or leave it alone.
727 		 * There may by stealth EXIT_DEAD tasks on ->children,
728 		 * forget_original_parent() must move them somewhere.
729 		 */
730 		pid_ns->child_reaper = init_pid_ns.child_reaper;
731 	}
732 
733 	return pid_ns->child_reaper;
734 }
735 
736 /*
737 * Any that need to be release_task'd are put on the @dead list.
738  */
739 static void reparent_leader(struct task_struct *father, struct task_struct *p,
740 				struct list_head *dead)
741 {
742 	list_move_tail(&p->sibling, &p->real_parent->children);
743 
744 	if (task_detached(p))
745 		return;
746 	/*
747 	 * If this is a threaded reparent there is no need to
748 	 * notify anyone anything has happened.
749 	 */
750 	if (same_thread_group(p->real_parent, father))
751 		return;
752 
753 	/* We don't want people slaying init.  */
754 	p->exit_signal = SIGCHLD;
755 
756 	/* If it has exited notify the new parent about this child's death. */
757 	if (!task_ptrace(p) &&
758 	    p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
759 		do_notify_parent(p, p->exit_signal);
760 		if (task_detached(p)) {
761 			p->exit_state = EXIT_DEAD;
762 			list_move_tail(&p->sibling, dead);
763 		}
764 	}
765 
766 	kill_orphaned_pgrp(p, father);
767 }
768 
769 static void forget_original_parent(struct task_struct *father)
770 {
771 	struct task_struct *p, *n, *reaper;
772 	LIST_HEAD(dead_children);
773 
774 	exit_ptrace(father);
775 
776 	write_lock_irq(&tasklist_lock);
777 	reaper = find_new_reaper(father);
778 
779 	list_for_each_entry_safe(p, n, &father->children, sibling) {
780 		struct task_struct *t = p;
781 		do {
782 			t->real_parent = reaper;
783 			if (t->parent == father) {
784 				BUG_ON(task_ptrace(t));
785 				t->parent = t->real_parent;
786 			}
787 			if (t->pdeath_signal)
788 				group_send_sig_info(t->pdeath_signal,
789 						    SEND_SIG_NOINFO, t);
790 		} while_each_thread(p, t);
791 		reparent_leader(father, p, &dead_children);
792 	}
793 	write_unlock_irq(&tasklist_lock);
794 
795 	BUG_ON(!list_empty(&father->children));
796 
797 	list_for_each_entry_safe(p, n, &dead_children, sibling) {
798 		list_del_init(&p->sibling);
799 		release_task(p);
800 	}
801 }
802 
803 /*
804  * Send signals to all our closest relatives so that they know
805  * to properly mourn us..
806  */
807 static void exit_notify(struct task_struct *tsk, int group_dead)
808 {
809 	int signal;
810 	void *cookie;
811 
812 	/*
813 	 * This does two things:
814 	 *
815   	 * A.  Make init inherit all the child processes
816 	 * B.  Check to see if any process groups have become orphaned
817 	 *	as a result of our exiting, and if they have any stopped
818 	 *	jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
819 	 */
820 	forget_original_parent(tsk);
821 	exit_task_namespaces(tsk);
822 
823 	write_lock_irq(&tasklist_lock);
824 	if (group_dead)
825 		kill_orphaned_pgrp(tsk->group_leader, NULL);
826 
827 	/* Let father know we died
828 	 *
829 	 * Thread signals are configurable, but you aren't going to use
830 	 * that to send signals to arbitary processes.
831 	 * That stops right now.
832 	 *
833 	 * If the parent exec id doesn't match the exec id we saved
834 	 * when we started then we know the parent has changed security
835 	 * domain.
836 	 *
837 	 * If our self_exec id doesn't match our parent_exec_id then
838 	 * we have changed execution domain as these two values started
839 	 * the same after a fork.
840 	 */
841 	if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
842 	    (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
843 	     tsk->self_exec_id != tsk->parent_exec_id))
844 		tsk->exit_signal = SIGCHLD;
845 
846 	signal = tracehook_notify_death(tsk, &cookie, group_dead);
847 	if (signal >= 0)
848 		signal = do_notify_parent(tsk, signal);
849 
850 	tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
851 
852 	/* mt-exec, de_thread() is waiting for us */
853 	if (thread_group_leader(tsk) &&
854 	    tsk->signal->group_exit_task &&
855 	    tsk->signal->notify_count < 0)
856 		wake_up_process(tsk->signal->group_exit_task);
857 
858 	write_unlock_irq(&tasklist_lock);
859 
860 	tracehook_report_death(tsk, signal, cookie, group_dead);
861 
862 	/* If the process is dead, release it - nobody will wait for it */
863 	if (signal == DEATH_REAP)
864 		release_task(tsk);
865 }
866 
867 #ifdef CONFIG_DEBUG_STACK_USAGE
868 static void check_stack_usage(void)
869 {
870 	static DEFINE_SPINLOCK(low_water_lock);
871 	static int lowest_to_date = THREAD_SIZE;
872 	unsigned long free;
873 
874 	free = stack_not_used(current);
875 
876 	if (free >= lowest_to_date)
877 		return;
878 
879 	spin_lock(&low_water_lock);
880 	if (free < lowest_to_date) {
881 		printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
882 				"left\n",
883 				current->comm, free);
884 		lowest_to_date = free;
885 	}
886 	spin_unlock(&low_water_lock);
887 }
888 #else
889 static inline void check_stack_usage(void) {}
890 #endif
891 
892 NORET_TYPE void do_exit(long code)
893 {
894 	struct task_struct *tsk = current;
895 	int group_dead;
896 
897 	profile_task_exit(tsk);
898 
899 	WARN_ON(atomic_read(&tsk->fs_excl));
900 
901 	if (unlikely(in_interrupt()))
902 		panic("Aiee, killing interrupt handler!");
903 	if (unlikely(!tsk->pid))
904 		panic("Attempted to kill the idle task!");
905 
906 	tracehook_report_exit(&code);
907 
908 	validate_creds_for_do_exit(tsk);
909 
910 	/*
911 	 * We're taking recursive faults here in do_exit. Safest is to just
912 	 * leave this task alone and wait for reboot.
913 	 */
914 	if (unlikely(tsk->flags & PF_EXITING)) {
915 		printk(KERN_ALERT
916 			"Fixing recursive fault but reboot is needed!\n");
917 		/*
918 		 * We can do this unlocked here. The futex code uses
919 		 * this flag just to verify whether the pi state
920 		 * cleanup has been done or not. In the worst case it
921 		 * loops once more. We pretend that the cleanup was
922 		 * done as there is no way to return. Either the
923 		 * OWNER_DIED bit is set by now or we push the blocked
924 		 * task into the wait for ever nirwana as well.
925 		 */
926 		tsk->flags |= PF_EXITPIDONE;
927 		set_current_state(TASK_UNINTERRUPTIBLE);
928 		schedule();
929 	}
930 
931 	exit_irq_thread();
932 
933 	exit_signals(tsk);  /* sets PF_EXITING */
934 	/*
935 	 * tsk->flags are checked in the futex code to protect against
936 	 * an exiting task cleaning up the robust pi futexes.
937 	 */
938 	smp_mb();
939 	raw_spin_unlock_wait(&tsk->pi_lock);
940 
941 	if (unlikely(in_atomic()))
942 		printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
943 				current->comm, task_pid_nr(current),
944 				preempt_count());
945 
946 	acct_update_integrals(tsk);
947 
948 	group_dead = atomic_dec_and_test(&tsk->signal->live);
949 	if (group_dead) {
950 		hrtimer_cancel(&tsk->signal->real_timer);
951 		exit_itimers(tsk->signal);
952 		if (tsk->mm)
953 			setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
954 	}
955 	acct_collect(code, group_dead);
956 	if (group_dead)
957 		tty_audit_exit();
958 	if (unlikely(tsk->audit_context))
959 		audit_free(tsk);
960 
961 	tsk->exit_code = code;
962 	taskstats_exit(tsk, group_dead);
963 
964 	exit_mm(tsk);
965 
966 	if (group_dead)
967 		acct_process();
968 	trace_sched_process_exit(tsk);
969 
970 	exit_sem(tsk);
971 	exit_files(tsk);
972 	exit_fs(tsk);
973 	check_stack_usage();
974 	exit_thread();
975 	cgroup_exit(tsk, 1);
976 
977 	if (group_dead)
978 		disassociate_ctty(1);
979 
980 	module_put(task_thread_info(tsk)->exec_domain->module);
981 
982 	proc_exit_connector(tsk);
983 
984 	/*
985 	 * FIXME: do that only when needed, using sched_exit tracepoint
986 	 */
987 	flush_ptrace_hw_breakpoint(tsk);
988 	/*
989 	 * Flush inherited counters to the parent - before the parent
990 	 * gets woken up by child-exit notifications.
991 	 */
992 	perf_event_exit_task(tsk);
993 
994 	exit_notify(tsk, group_dead);
995 #ifdef CONFIG_NUMA
996 	mpol_put(tsk->mempolicy);
997 	tsk->mempolicy = NULL;
998 #endif
999 #ifdef CONFIG_FUTEX
1000 	if (unlikely(current->pi_state_cache))
1001 		kfree(current->pi_state_cache);
1002 #endif
1003 	/*
1004 	 * Make sure we are holding no locks:
1005 	 */
1006 	debug_check_no_locks_held(tsk);
1007 	/*
1008 	 * We can do this unlocked here. The futex code uses this flag
1009 	 * just to verify whether the pi state cleanup has been done
1010 	 * or not. In the worst case it loops once more.
1011 	 */
1012 	tsk->flags |= PF_EXITPIDONE;
1013 
1014 	if (tsk->io_context)
1015 		exit_io_context(tsk);
1016 
1017 	if (tsk->splice_pipe)
1018 		__free_pipe_info(tsk->splice_pipe);
1019 
1020 	validate_creds_for_do_exit(tsk);
1021 
1022 	preempt_disable();
1023 	exit_rcu();
1024 	/* causes final put_task_struct in finish_task_switch(). */
1025 	tsk->state = TASK_DEAD;
1026 	schedule();
1027 	BUG();
1028 	/* Avoid "noreturn function does return".  */
1029 	for (;;)
1030 		cpu_relax();	/* For when BUG is null */
1031 }
1032 
1033 EXPORT_SYMBOL_GPL(do_exit);
1034 
1035 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1036 {
1037 	if (comp)
1038 		complete(comp);
1039 
1040 	do_exit(code);
1041 }
1042 
1043 EXPORT_SYMBOL(complete_and_exit);
1044 
1045 SYSCALL_DEFINE1(exit, int, error_code)
1046 {
1047 	do_exit((error_code&0xff)<<8);
1048 }
1049 
1050 /*
1051  * Take down every thread in the group.  This is called by fatal signals
1052  * as well as by sys_exit_group (below).
1053  */
1054 NORET_TYPE void
1055 do_group_exit(int exit_code)
1056 {
1057 	struct signal_struct *sig = current->signal;
1058 
1059 	BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1060 
1061 	if (signal_group_exit(sig))
1062 		exit_code = sig->group_exit_code;
1063 	else if (!thread_group_empty(current)) {
1064 		struct sighand_struct *const sighand = current->sighand;
1065 		spin_lock_irq(&sighand->siglock);
1066 		if (signal_group_exit(sig))
1067 			/* Another thread got here before we took the lock.  */
1068 			exit_code = sig->group_exit_code;
1069 		else {
1070 			sig->group_exit_code = exit_code;
1071 			sig->flags = SIGNAL_GROUP_EXIT;
1072 			zap_other_threads(current);
1073 		}
1074 		spin_unlock_irq(&sighand->siglock);
1075 	}
1076 
1077 	do_exit(exit_code);
1078 	/* NOTREACHED */
1079 }
1080 
1081 /*
1082  * this kills every thread in the thread group. Note that any externally
1083  * wait4()-ing process will get the correct exit code - even if this
1084  * thread is not the thread group leader.
1085  */
1086 SYSCALL_DEFINE1(exit_group, int, error_code)
1087 {
1088 	do_group_exit((error_code & 0xff) << 8);
1089 	/* NOTREACHED */
1090 	return 0;
1091 }
1092 
1093 struct wait_opts {
1094 	enum pid_type		wo_type;
1095 	int			wo_flags;
1096 	struct pid		*wo_pid;
1097 
1098 	struct siginfo __user	*wo_info;
1099 	int __user		*wo_stat;
1100 	struct rusage __user	*wo_rusage;
1101 
1102 	wait_queue_t		child_wait;
1103 	int			notask_error;
1104 };
1105 
1106 static inline
1107 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1108 {
1109 	if (type != PIDTYPE_PID)
1110 		task = task->group_leader;
1111 	return task->pids[type].pid;
1112 }
1113 
1114 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1115 {
1116 	return	wo->wo_type == PIDTYPE_MAX ||
1117 		task_pid_type(p, wo->wo_type) == wo->wo_pid;
1118 }
1119 
1120 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1121 {
1122 	if (!eligible_pid(wo, p))
1123 		return 0;
1124 	/* Wait for all children (clone and not) if __WALL is set;
1125 	 * otherwise, wait for clone children *only* if __WCLONE is
1126 	 * set; otherwise, wait for non-clone children *only*.  (Note:
1127 	 * A "clone" child here is one that reports to its parent
1128 	 * using a signal other than SIGCHLD.) */
1129 	if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1130 	    && !(wo->wo_flags & __WALL))
1131 		return 0;
1132 
1133 	return 1;
1134 }
1135 
1136 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1137 				pid_t pid, uid_t uid, int why, int status)
1138 {
1139 	struct siginfo __user *infop;
1140 	int retval = wo->wo_rusage
1141 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1142 
1143 	put_task_struct(p);
1144 	infop = wo->wo_info;
1145 	if (infop) {
1146 		if (!retval)
1147 			retval = put_user(SIGCHLD, &infop->si_signo);
1148 		if (!retval)
1149 			retval = put_user(0, &infop->si_errno);
1150 		if (!retval)
1151 			retval = put_user((short)why, &infop->si_code);
1152 		if (!retval)
1153 			retval = put_user(pid, &infop->si_pid);
1154 		if (!retval)
1155 			retval = put_user(uid, &infop->si_uid);
1156 		if (!retval)
1157 			retval = put_user(status, &infop->si_status);
1158 	}
1159 	if (!retval)
1160 		retval = pid;
1161 	return retval;
1162 }
1163 
1164 /*
1165  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
1166  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1167  * the lock and this task is uninteresting.  If we return nonzero, we have
1168  * released the lock and the system call should return.
1169  */
1170 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1171 {
1172 	unsigned long state;
1173 	int retval, status, traced;
1174 	pid_t pid = task_pid_vnr(p);
1175 	uid_t uid = __task_cred(p)->uid;
1176 	struct siginfo __user *infop;
1177 
1178 	if (!likely(wo->wo_flags & WEXITED))
1179 		return 0;
1180 
1181 	if (unlikely(wo->wo_flags & WNOWAIT)) {
1182 		int exit_code = p->exit_code;
1183 		int why, status;
1184 
1185 		get_task_struct(p);
1186 		read_unlock(&tasklist_lock);
1187 		if ((exit_code & 0x7f) == 0) {
1188 			why = CLD_EXITED;
1189 			status = exit_code >> 8;
1190 		} else {
1191 			why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1192 			status = exit_code & 0x7f;
1193 		}
1194 		return wait_noreap_copyout(wo, p, pid, uid, why, status);
1195 	}
1196 
1197 	/*
1198 	 * Try to move the task's state to DEAD
1199 	 * only one thread is allowed to do this:
1200 	 */
1201 	state = xchg(&p->exit_state, EXIT_DEAD);
1202 	if (state != EXIT_ZOMBIE) {
1203 		BUG_ON(state != EXIT_DEAD);
1204 		return 0;
1205 	}
1206 
1207 	traced = ptrace_reparented(p);
1208 	/*
1209 	 * It can be ptraced but not reparented, check
1210 	 * !task_detached() to filter out sub-threads.
1211 	 */
1212 	if (likely(!traced) && likely(!task_detached(p))) {
1213 		struct signal_struct *psig;
1214 		struct signal_struct *sig;
1215 		unsigned long maxrss;
1216 		cputime_t tgutime, tgstime;
1217 
1218 		/*
1219 		 * The resource counters for the group leader are in its
1220 		 * own task_struct.  Those for dead threads in the group
1221 		 * are in its signal_struct, as are those for the child
1222 		 * processes it has previously reaped.  All these
1223 		 * accumulate in the parent's signal_struct c* fields.
1224 		 *
1225 		 * We don't bother to take a lock here to protect these
1226 		 * p->signal fields, because they are only touched by
1227 		 * __exit_signal, which runs with tasklist_lock
1228 		 * write-locked anyway, and so is excluded here.  We do
1229 		 * need to protect the access to parent->signal fields,
1230 		 * as other threads in the parent group can be right
1231 		 * here reaping other children at the same time.
1232 		 *
1233 		 * We use thread_group_times() to get times for the thread
1234 		 * group, which consolidates times for all threads in the
1235 		 * group including the group leader.
1236 		 */
1237 		thread_group_times(p, &tgutime, &tgstime);
1238 		spin_lock_irq(&p->real_parent->sighand->siglock);
1239 		psig = p->real_parent->signal;
1240 		sig = p->signal;
1241 		psig->cutime =
1242 			cputime_add(psig->cutime,
1243 			cputime_add(tgutime,
1244 				    sig->cutime));
1245 		psig->cstime =
1246 			cputime_add(psig->cstime,
1247 			cputime_add(tgstime,
1248 				    sig->cstime));
1249 		psig->cgtime =
1250 			cputime_add(psig->cgtime,
1251 			cputime_add(p->gtime,
1252 			cputime_add(sig->gtime,
1253 				    sig->cgtime)));
1254 		psig->cmin_flt +=
1255 			p->min_flt + sig->min_flt + sig->cmin_flt;
1256 		psig->cmaj_flt +=
1257 			p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1258 		psig->cnvcsw +=
1259 			p->nvcsw + sig->nvcsw + sig->cnvcsw;
1260 		psig->cnivcsw +=
1261 			p->nivcsw + sig->nivcsw + sig->cnivcsw;
1262 		psig->cinblock +=
1263 			task_io_get_inblock(p) +
1264 			sig->inblock + sig->cinblock;
1265 		psig->coublock +=
1266 			task_io_get_oublock(p) +
1267 			sig->oublock + sig->coublock;
1268 		maxrss = max(sig->maxrss, sig->cmaxrss);
1269 		if (psig->cmaxrss < maxrss)
1270 			psig->cmaxrss = maxrss;
1271 		task_io_accounting_add(&psig->ioac, &p->ioac);
1272 		task_io_accounting_add(&psig->ioac, &sig->ioac);
1273 		spin_unlock_irq(&p->real_parent->sighand->siglock);
1274 	}
1275 
1276 	/*
1277 	 * Now we are sure this task is interesting, and no other
1278 	 * thread can reap it because we set its state to EXIT_DEAD.
1279 	 */
1280 	read_unlock(&tasklist_lock);
1281 
1282 	retval = wo->wo_rusage
1283 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1284 	status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1285 		? p->signal->group_exit_code : p->exit_code;
1286 	if (!retval && wo->wo_stat)
1287 		retval = put_user(status, wo->wo_stat);
1288 
1289 	infop = wo->wo_info;
1290 	if (!retval && infop)
1291 		retval = put_user(SIGCHLD, &infop->si_signo);
1292 	if (!retval && infop)
1293 		retval = put_user(0, &infop->si_errno);
1294 	if (!retval && infop) {
1295 		int why;
1296 
1297 		if ((status & 0x7f) == 0) {
1298 			why = CLD_EXITED;
1299 			status >>= 8;
1300 		} else {
1301 			why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1302 			status &= 0x7f;
1303 		}
1304 		retval = put_user((short)why, &infop->si_code);
1305 		if (!retval)
1306 			retval = put_user(status, &infop->si_status);
1307 	}
1308 	if (!retval && infop)
1309 		retval = put_user(pid, &infop->si_pid);
1310 	if (!retval && infop)
1311 		retval = put_user(uid, &infop->si_uid);
1312 	if (!retval)
1313 		retval = pid;
1314 
1315 	if (traced) {
1316 		write_lock_irq(&tasklist_lock);
1317 		/* We dropped tasklist, ptracer could die and untrace */
1318 		ptrace_unlink(p);
1319 		/*
1320 		 * If this is not a detached task, notify the parent.
1321 		 * If it's still not detached after that, don't release
1322 		 * it now.
1323 		 */
1324 		if (!task_detached(p)) {
1325 			do_notify_parent(p, p->exit_signal);
1326 			if (!task_detached(p)) {
1327 				p->exit_state = EXIT_ZOMBIE;
1328 				p = NULL;
1329 			}
1330 		}
1331 		write_unlock_irq(&tasklist_lock);
1332 	}
1333 	if (p != NULL)
1334 		release_task(p);
1335 
1336 	return retval;
1337 }
1338 
1339 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1340 {
1341 	if (ptrace) {
1342 		if (task_is_stopped_or_traced(p))
1343 			return &p->exit_code;
1344 	} else {
1345 		if (p->signal->flags & SIGNAL_STOP_STOPPED)
1346 			return &p->signal->group_exit_code;
1347 	}
1348 	return NULL;
1349 }
1350 
1351 /*
1352  * Handle sys_wait4 work for one task in state TASK_STOPPED.  We hold
1353  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1354  * the lock and this task is uninteresting.  If we return nonzero, we have
1355  * released the lock and the system call should return.
1356  */
1357 static int wait_task_stopped(struct wait_opts *wo,
1358 				int ptrace, struct task_struct *p)
1359 {
1360 	struct siginfo __user *infop;
1361 	int retval, exit_code, *p_code, why;
1362 	uid_t uid = 0; /* unneeded, required by compiler */
1363 	pid_t pid;
1364 
1365 	/*
1366 	 * Traditionally we see ptrace'd stopped tasks regardless of options.
1367 	 */
1368 	if (!ptrace && !(wo->wo_flags & WUNTRACED))
1369 		return 0;
1370 
1371 	exit_code = 0;
1372 	spin_lock_irq(&p->sighand->siglock);
1373 
1374 	p_code = task_stopped_code(p, ptrace);
1375 	if (unlikely(!p_code))
1376 		goto unlock_sig;
1377 
1378 	exit_code = *p_code;
1379 	if (!exit_code)
1380 		goto unlock_sig;
1381 
1382 	if (!unlikely(wo->wo_flags & WNOWAIT))
1383 		*p_code = 0;
1384 
1385 	/* don't need the RCU readlock here as we're holding a spinlock */
1386 	uid = __task_cred(p)->uid;
1387 unlock_sig:
1388 	spin_unlock_irq(&p->sighand->siglock);
1389 	if (!exit_code)
1390 		return 0;
1391 
1392 	/*
1393 	 * Now we are pretty sure this task is interesting.
1394 	 * Make sure it doesn't get reaped out from under us while we
1395 	 * give up the lock and then examine it below.  We don't want to
1396 	 * keep holding onto the tasklist_lock while we call getrusage and
1397 	 * possibly take page faults for user memory.
1398 	 */
1399 	get_task_struct(p);
1400 	pid = task_pid_vnr(p);
1401 	why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1402 	read_unlock(&tasklist_lock);
1403 
1404 	if (unlikely(wo->wo_flags & WNOWAIT))
1405 		return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1406 
1407 	retval = wo->wo_rusage
1408 		? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1409 	if (!retval && wo->wo_stat)
1410 		retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1411 
1412 	infop = wo->wo_info;
1413 	if (!retval && infop)
1414 		retval = put_user(SIGCHLD, &infop->si_signo);
1415 	if (!retval && infop)
1416 		retval = put_user(0, &infop->si_errno);
1417 	if (!retval && infop)
1418 		retval = put_user((short)why, &infop->si_code);
1419 	if (!retval && infop)
1420 		retval = put_user(exit_code, &infop->si_status);
1421 	if (!retval && infop)
1422 		retval = put_user(pid, &infop->si_pid);
1423 	if (!retval && infop)
1424 		retval = put_user(uid, &infop->si_uid);
1425 	if (!retval)
1426 		retval = pid;
1427 	put_task_struct(p);
1428 
1429 	BUG_ON(!retval);
1430 	return retval;
1431 }
1432 
1433 /*
1434  * Handle do_wait work for one task in a live, non-stopped state.
1435  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1436  * the lock and this task is uninteresting.  If we return nonzero, we have
1437  * released the lock and the system call should return.
1438  */
1439 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1440 {
1441 	int retval;
1442 	pid_t pid;
1443 	uid_t uid;
1444 
1445 	if (!unlikely(wo->wo_flags & WCONTINUED))
1446 		return 0;
1447 
1448 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1449 		return 0;
1450 
1451 	spin_lock_irq(&p->sighand->siglock);
1452 	/* Re-check with the lock held.  */
1453 	if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1454 		spin_unlock_irq(&p->sighand->siglock);
1455 		return 0;
1456 	}
1457 	if (!unlikely(wo->wo_flags & WNOWAIT))
1458 		p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1459 	uid = __task_cred(p)->uid;
1460 	spin_unlock_irq(&p->sighand->siglock);
1461 
1462 	pid = task_pid_vnr(p);
1463 	get_task_struct(p);
1464 	read_unlock(&tasklist_lock);
1465 
1466 	if (!wo->wo_info) {
1467 		retval = wo->wo_rusage
1468 			? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1469 		put_task_struct(p);
1470 		if (!retval && wo->wo_stat)
1471 			retval = put_user(0xffff, wo->wo_stat);
1472 		if (!retval)
1473 			retval = pid;
1474 	} else {
1475 		retval = wait_noreap_copyout(wo, p, pid, uid,
1476 					     CLD_CONTINUED, SIGCONT);
1477 		BUG_ON(retval == 0);
1478 	}
1479 
1480 	return retval;
1481 }
1482 
1483 /*
1484  * Consider @p for a wait by @parent.
1485  *
1486  * -ECHILD should be in ->notask_error before the first call.
1487  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1488  * Returns zero if the search for a child should continue;
1489  * then ->notask_error is 0 if @p is an eligible child,
1490  * or another error from security_task_wait(), or still -ECHILD.
1491  */
1492 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1493 				struct task_struct *p)
1494 {
1495 	int ret = eligible_child(wo, p);
1496 	if (!ret)
1497 		return ret;
1498 
1499 	ret = security_task_wait(p);
1500 	if (unlikely(ret < 0)) {
1501 		/*
1502 		 * If we have not yet seen any eligible child,
1503 		 * then let this error code replace -ECHILD.
1504 		 * A permission error will give the user a clue
1505 		 * to look for security policy problems, rather
1506 		 * than for mysterious wait bugs.
1507 		 */
1508 		if (wo->notask_error)
1509 			wo->notask_error = ret;
1510 		return 0;
1511 	}
1512 
1513 	if (likely(!ptrace) && unlikely(task_ptrace(p))) {
1514 		/*
1515 		 * This child is hidden by ptrace.
1516 		 * We aren't allowed to see it now, but eventually we will.
1517 		 */
1518 		wo->notask_error = 0;
1519 		return 0;
1520 	}
1521 
1522 	if (p->exit_state == EXIT_DEAD)
1523 		return 0;
1524 
1525 	/*
1526 	 * We don't reap group leaders with subthreads.
1527 	 */
1528 	if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1529 		return wait_task_zombie(wo, p);
1530 
1531 	/*
1532 	 * It's stopped or running now, so it might
1533 	 * later continue, exit, or stop again.
1534 	 */
1535 	wo->notask_error = 0;
1536 
1537 	if (task_stopped_code(p, ptrace))
1538 		return wait_task_stopped(wo, ptrace, p);
1539 
1540 	return wait_task_continued(wo, p);
1541 }
1542 
1543 /*
1544  * Do the work of do_wait() for one thread in the group, @tsk.
1545  *
1546  * -ECHILD should be in ->notask_error before the first call.
1547  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1548  * Returns zero if the search for a child should continue; then
1549  * ->notask_error is 0 if there were any eligible children,
1550  * or another error from security_task_wait(), or still -ECHILD.
1551  */
1552 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1553 {
1554 	struct task_struct *p;
1555 
1556 	list_for_each_entry(p, &tsk->children, sibling) {
1557 		int ret = wait_consider_task(wo, 0, p);
1558 		if (ret)
1559 			return ret;
1560 	}
1561 
1562 	return 0;
1563 }
1564 
1565 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1566 {
1567 	struct task_struct *p;
1568 
1569 	list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1570 		int ret = wait_consider_task(wo, 1, p);
1571 		if (ret)
1572 			return ret;
1573 	}
1574 
1575 	return 0;
1576 }
1577 
1578 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1579 				int sync, void *key)
1580 {
1581 	struct wait_opts *wo = container_of(wait, struct wait_opts,
1582 						child_wait);
1583 	struct task_struct *p = key;
1584 
1585 	if (!eligible_pid(wo, p))
1586 		return 0;
1587 
1588 	if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1589 		return 0;
1590 
1591 	return default_wake_function(wait, mode, sync, key);
1592 }
1593 
1594 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1595 {
1596 	__wake_up_sync_key(&parent->signal->wait_chldexit,
1597 				TASK_INTERRUPTIBLE, 1, p);
1598 }
1599 
1600 static long do_wait(struct wait_opts *wo)
1601 {
1602 	struct task_struct *tsk;
1603 	int retval;
1604 
1605 	trace_sched_process_wait(wo->wo_pid);
1606 
1607 	init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1608 	wo->child_wait.private = current;
1609 	add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1610 repeat:
1611 	/*
1612 	 * If there is nothing that can match our critiera just get out.
1613 	 * We will clear ->notask_error to zero if we see any child that
1614 	 * might later match our criteria, even if we are not able to reap
1615 	 * it yet.
1616 	 */
1617 	wo->notask_error = -ECHILD;
1618 	if ((wo->wo_type < PIDTYPE_MAX) &&
1619 	   (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1620 		goto notask;
1621 
1622 	set_current_state(TASK_INTERRUPTIBLE);
1623 	read_lock(&tasklist_lock);
1624 	tsk = current;
1625 	do {
1626 		retval = do_wait_thread(wo, tsk);
1627 		if (retval)
1628 			goto end;
1629 
1630 		retval = ptrace_do_wait(wo, tsk);
1631 		if (retval)
1632 			goto end;
1633 
1634 		if (wo->wo_flags & __WNOTHREAD)
1635 			break;
1636 	} while_each_thread(current, tsk);
1637 	read_unlock(&tasklist_lock);
1638 
1639 notask:
1640 	retval = wo->notask_error;
1641 	if (!retval && !(wo->wo_flags & WNOHANG)) {
1642 		retval = -ERESTARTSYS;
1643 		if (!signal_pending(current)) {
1644 			schedule();
1645 			goto repeat;
1646 		}
1647 	}
1648 end:
1649 	__set_current_state(TASK_RUNNING);
1650 	remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1651 	return retval;
1652 }
1653 
1654 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1655 		infop, int, options, struct rusage __user *, ru)
1656 {
1657 	struct wait_opts wo;
1658 	struct pid *pid = NULL;
1659 	enum pid_type type;
1660 	long ret;
1661 
1662 	if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1663 		return -EINVAL;
1664 	if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1665 		return -EINVAL;
1666 
1667 	switch (which) {
1668 	case P_ALL:
1669 		type = PIDTYPE_MAX;
1670 		break;
1671 	case P_PID:
1672 		type = PIDTYPE_PID;
1673 		if (upid <= 0)
1674 			return -EINVAL;
1675 		break;
1676 	case P_PGID:
1677 		type = PIDTYPE_PGID;
1678 		if (upid <= 0)
1679 			return -EINVAL;
1680 		break;
1681 	default:
1682 		return -EINVAL;
1683 	}
1684 
1685 	if (type < PIDTYPE_MAX)
1686 		pid = find_get_pid(upid);
1687 
1688 	wo.wo_type	= type;
1689 	wo.wo_pid	= pid;
1690 	wo.wo_flags	= options;
1691 	wo.wo_info	= infop;
1692 	wo.wo_stat	= NULL;
1693 	wo.wo_rusage	= ru;
1694 	ret = do_wait(&wo);
1695 
1696 	if (ret > 0) {
1697 		ret = 0;
1698 	} else if (infop) {
1699 		/*
1700 		 * For a WNOHANG return, clear out all the fields
1701 		 * we would set so the user can easily tell the
1702 		 * difference.
1703 		 */
1704 		if (!ret)
1705 			ret = put_user(0, &infop->si_signo);
1706 		if (!ret)
1707 			ret = put_user(0, &infop->si_errno);
1708 		if (!ret)
1709 			ret = put_user(0, &infop->si_code);
1710 		if (!ret)
1711 			ret = put_user(0, &infop->si_pid);
1712 		if (!ret)
1713 			ret = put_user(0, &infop->si_uid);
1714 		if (!ret)
1715 			ret = put_user(0, &infop->si_status);
1716 	}
1717 
1718 	put_pid(pid);
1719 
1720 	/* avoid REGPARM breakage on x86: */
1721 	asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1722 	return ret;
1723 }
1724 
1725 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1726 		int, options, struct rusage __user *, ru)
1727 {
1728 	struct wait_opts wo;
1729 	struct pid *pid = NULL;
1730 	enum pid_type type;
1731 	long ret;
1732 
1733 	if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1734 			__WNOTHREAD|__WCLONE|__WALL))
1735 		return -EINVAL;
1736 
1737 	if (upid == -1)
1738 		type = PIDTYPE_MAX;
1739 	else if (upid < 0) {
1740 		type = PIDTYPE_PGID;
1741 		pid = find_get_pid(-upid);
1742 	} else if (upid == 0) {
1743 		type = PIDTYPE_PGID;
1744 		pid = get_task_pid(current, PIDTYPE_PGID);
1745 	} else /* upid > 0 */ {
1746 		type = PIDTYPE_PID;
1747 		pid = find_get_pid(upid);
1748 	}
1749 
1750 	wo.wo_type	= type;
1751 	wo.wo_pid	= pid;
1752 	wo.wo_flags	= options | WEXITED;
1753 	wo.wo_info	= NULL;
1754 	wo.wo_stat	= stat_addr;
1755 	wo.wo_rusage	= ru;
1756 	ret = do_wait(&wo);
1757 	put_pid(pid);
1758 
1759 	/* avoid REGPARM breakage on x86: */
1760 	asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1761 	return ret;
1762 }
1763 
1764 #ifdef __ARCH_WANT_SYS_WAITPID
1765 
1766 /*
1767  * sys_waitpid() remains for compatibility. waitpid() should be
1768  * implemented by calling sys_wait4() from libc.a.
1769  */
1770 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1771 {
1772 	return sys_wait4(pid, stat_addr, options, NULL);
1773 }
1774 
1775 #endif
1776