1f236ed375f83c2bf6b3c88c65d94d6ac9b2c0b3
[linux-2.6-microblaze.git] / kernel / exit.c
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 *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         seccomp_filter_release(p);
221         proc_flush_pid(thread_pid);
222         put_pid(thread_pid);
223         release_thread(p);
224         put_task_struct_rcu_user(p);
225
226         p = leader;
227         if (unlikely(zap_leader))
228                 goto repeat;
229 }
230
231 int rcuwait_wake_up(struct rcuwait *w)
232 {
233         int ret = 0;
234         struct task_struct *task;
235
236         rcu_read_lock();
237
238         /*
239          * Order condition vs @task, such that everything prior to the load
240          * of @task is visible. This is the condition as to why the user called
241          * rcuwait_wake() in the first place. Pairs with set_current_state()
242          * barrier (A) in rcuwait_wait_event().
243          *
244          *    WAIT                WAKE
245          *    [S] tsk = current   [S] cond = true
246          *        MB (A)              MB (B)
247          *    [L] cond            [L] tsk
248          */
249         smp_mb(); /* (B) */
250
251         task = rcu_dereference(w->task);
252         if (task)
253                 ret = wake_up_process(task);
254         rcu_read_unlock();
255
256         return ret;
257 }
258 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
259
260 /*
261  * Determine if a process group is "orphaned", according to the POSIX
262  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
263  * by terminal-generated stop signals.  Newly orphaned process groups are
264  * to receive a SIGHUP and a SIGCONT.
265  *
266  * "I ask you, have you ever known what it is to be an orphan?"
267  */
268 static int will_become_orphaned_pgrp(struct pid *pgrp,
269                                         struct task_struct *ignored_task)
270 {
271         struct task_struct *p;
272
273         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
274                 if ((p == ignored_task) ||
275                     (p->exit_state && thread_group_empty(p)) ||
276                     is_global_init(p->real_parent))
277                         continue;
278
279                 if (task_pgrp(p->real_parent) != pgrp &&
280                     task_session(p->real_parent) == task_session(p))
281                         return 0;
282         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
283
284         return 1;
285 }
286
287 int is_current_pgrp_orphaned(void)
288 {
289         int retval;
290
291         read_lock(&tasklist_lock);
292         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
293         read_unlock(&tasklist_lock);
294
295         return retval;
296 }
297
298 static bool has_stopped_jobs(struct pid *pgrp)
299 {
300         struct task_struct *p;
301
302         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
303                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
304                         return true;
305         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
306
307         return false;
308 }
309
310 /*
311  * Check to see if any process groups have become orphaned as
312  * a result of our exiting, and if they have any stopped jobs,
313  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
314  */
315 static void
316 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
317 {
318         struct pid *pgrp = task_pgrp(tsk);
319         struct task_struct *ignored_task = tsk;
320
321         if (!parent)
322                 /* exit: our father is in a different pgrp than
323                  * we are and we were the only connection outside.
324                  */
325                 parent = tsk->real_parent;
326         else
327                 /* reparent: our child is in a different pgrp than
328                  * we are, and it was the only connection outside.
329                  */
330                 ignored_task = NULL;
331
332         if (task_pgrp(parent) != pgrp &&
333             task_session(parent) == task_session(tsk) &&
334             will_become_orphaned_pgrp(pgrp, ignored_task) &&
335             has_stopped_jobs(pgrp)) {
336                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
337                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
338         }
339 }
340
341 #ifdef CONFIG_MEMCG
342 /*
343  * A task is exiting.   If it owned this mm, find a new owner for the mm.
344  */
345 void mm_update_next_owner(struct mm_struct *mm)
346 {
347         struct task_struct *c, *g, *p = current;
348
349 retry:
350         /*
351          * If the exiting or execing task is not the owner, it's
352          * someone else's problem.
353          */
354         if (mm->owner != p)
355                 return;
356         /*
357          * The current owner is exiting/execing and there are no other
358          * candidates.  Do not leave the mm pointing to a possibly
359          * freed task structure.
360          */
361         if (atomic_read(&mm->mm_users) <= 1) {
362                 WRITE_ONCE(mm->owner, NULL);
363                 return;
364         }
365
366         read_lock(&tasklist_lock);
367         /*
368          * Search in the children
369          */
370         list_for_each_entry(c, &p->children, sibling) {
371                 if (c->mm == mm)
372                         goto assign_new_owner;
373         }
374
375         /*
376          * Search in the siblings
377          */
378         list_for_each_entry(c, &p->real_parent->children, sibling) {
379                 if (c->mm == mm)
380                         goto assign_new_owner;
381         }
382
383         /*
384          * Search through everything else, we should not get here often.
385          */
386         for_each_process(g) {
387                 if (g->flags & PF_KTHREAD)
388                         continue;
389                 for_each_thread(g, c) {
390                         if (c->mm == mm)
391                                 goto assign_new_owner;
392                         if (c->mm)
393                                 break;
394                 }
395         }
396         read_unlock(&tasklist_lock);
397         /*
398          * We found no owner yet mm_users > 1: this implies that we are
399          * most likely racing with swapoff (try_to_unuse()) or /proc or
400          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
401          */
402         WRITE_ONCE(mm->owner, NULL);
403         return;
404
405 assign_new_owner:
406         BUG_ON(c == p);
407         get_task_struct(c);
408         /*
409          * The task_lock protects c->mm from changing.
410          * We always want mm->owner->mm == mm
411          */
412         task_lock(c);
413         /*
414          * Delay read_unlock() till we have the task_lock()
415          * to ensure that c does not slip away underneath us
416          */
417         read_unlock(&tasklist_lock);
418         if (c->mm != mm) {
419                 task_unlock(c);
420                 put_task_struct(c);
421                 goto retry;
422         }
423         WRITE_ONCE(mm->owner, c);
424         task_unlock(c);
425         put_task_struct(c);
426 }
427 #endif /* CONFIG_MEMCG */
428
429 /*
430  * Turn us into a lazy TLB process if we
431  * aren't already..
432  */
433 static void exit_mm(void)
434 {
435         struct mm_struct *mm = current->mm;
436         struct core_state *core_state;
437
438         exit_mm_release(current, mm);
439         if (!mm)
440                 return;
441         sync_mm_rss(mm);
442         /*
443          * Serialize with any possible pending coredump.
444          * We must hold mmap_lock around checking core_state
445          * and clearing tsk->mm.  The core-inducing thread
446          * will increment ->nr_threads for each thread in the
447          * group with ->mm != NULL.
448          */
449         mmap_read_lock(mm);
450         core_state = mm->core_state;
451         if (core_state) {
452                 struct core_thread self;
453
454                 mmap_read_unlock(mm);
455
456                 self.task = current;
457                 if (self.task->flags & PF_SIGNALED)
458                         self.next = xchg(&core_state->dumper.next, &self);
459                 else
460                         self.task = NULL;
461                 /*
462                  * Implies mb(), the result of xchg() must be visible
463                  * to core_state->dumper.
464                  */
465                 if (atomic_dec_and_test(&core_state->nr_threads))
466                         complete(&core_state->startup);
467
468                 for (;;) {
469                         set_current_state(TASK_UNINTERRUPTIBLE);
470                         if (!self.task) /* see coredump_finish() */
471                                 break;
472                         freezable_schedule();
473                 }
474                 __set_current_state(TASK_RUNNING);
475                 mmap_read_lock(mm);
476         }
477         mmgrab(mm);
478         BUG_ON(mm != current->active_mm);
479         /* more a memory barrier than a real lock */
480         task_lock(current);
481         current->mm = NULL;
482         mmap_read_unlock(mm);
483         enter_lazy_tlb(mm, current);
484         task_unlock(current);
485         mm_update_next_owner(mm);
486         mmput(mm);
487         if (test_thread_flag(TIF_MEMDIE))
488                 exit_oom_victim();
489 }
490
491 static struct task_struct *find_alive_thread(struct task_struct *p)
492 {
493         struct task_struct *t;
494
495         for_each_thread(p, t) {
496                 if (!(t->flags & PF_EXITING))
497                         return t;
498         }
499         return NULL;
500 }
501
502 static struct task_struct *find_child_reaper(struct task_struct *father,
503                                                 struct list_head *dead)
504         __releases(&tasklist_lock)
505         __acquires(&tasklist_lock)
506 {
507         struct pid_namespace *pid_ns = task_active_pid_ns(father);
508         struct task_struct *reaper = pid_ns->child_reaper;
509         struct task_struct *p, *n;
510
511         if (likely(reaper != father))
512                 return reaper;
513
514         reaper = find_alive_thread(father);
515         if (reaper) {
516                 pid_ns->child_reaper = reaper;
517                 return reaper;
518         }
519
520         write_unlock_irq(&tasklist_lock);
521
522         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
523                 list_del_init(&p->ptrace_entry);
524                 release_task(p);
525         }
526
527         zap_pid_ns_processes(pid_ns);
528         write_lock_irq(&tasklist_lock);
529
530         return father;
531 }
532
533 /*
534  * When we die, we re-parent all our children, and try to:
535  * 1. give them to another thread in our thread group, if such a member exists
536  * 2. give it to the first ancestor process which prctl'd itself as a
537  *    child_subreaper for its children (like a service manager)
538  * 3. give it to the init process (PID 1) in our pid namespace
539  */
540 static struct task_struct *find_new_reaper(struct task_struct *father,
541                                            struct task_struct *child_reaper)
542 {
543         struct task_struct *thread, *reaper;
544
545         thread = find_alive_thread(father);
546         if (thread)
547                 return thread;
548
549         if (father->signal->has_child_subreaper) {
550                 unsigned int ns_level = task_pid(father)->level;
551                 /*
552                  * Find the first ->is_child_subreaper ancestor in our pid_ns.
553                  * We can't check reaper != child_reaper to ensure we do not
554                  * cross the namespaces, the exiting parent could be injected
555                  * by setns() + fork().
556                  * We check pid->level, this is slightly more efficient than
557                  * task_active_pid_ns(reaper) != task_active_pid_ns(father).
558                  */
559                 for (reaper = father->real_parent;
560                      task_pid(reaper)->level == ns_level;
561                      reaper = reaper->real_parent) {
562                         if (reaper == &init_task)
563                                 break;
564                         if (!reaper->signal->is_child_subreaper)
565                                 continue;
566                         thread = find_alive_thread(reaper);
567                         if (thread)
568                                 return thread;
569                 }
570         }
571
572         return child_reaper;
573 }
574
575 /*
576 * Any that need to be release_task'd are put on the @dead list.
577  */
578 static void reparent_leader(struct task_struct *father, struct task_struct *p,
579                                 struct list_head *dead)
580 {
581         if (unlikely(p->exit_state == EXIT_DEAD))
582                 return;
583
584         /* We don't want people slaying init. */
585         p->exit_signal = SIGCHLD;
586
587         /* If it has exited notify the new parent about this child's death. */
588         if (!p->ptrace &&
589             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
590                 if (do_notify_parent(p, p->exit_signal)) {
591                         p->exit_state = EXIT_DEAD;
592                         list_add(&p->ptrace_entry, dead);
593                 }
594         }
595
596         kill_orphaned_pgrp(p, father);
597 }
598
599 /*
600  * This does two things:
601  *
602  * A.  Make init inherit all the child processes
603  * B.  Check to see if any process groups have become orphaned
604  *      as a result of our exiting, and if they have any stopped
605  *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
606  */
607 static void forget_original_parent(struct task_struct *father,
608                                         struct list_head *dead)
609 {
610         struct task_struct *p, *t, *reaper;
611
612         if (unlikely(!list_empty(&father->ptraced)))
613                 exit_ptrace(father, dead);
614
615         /* Can drop and reacquire tasklist_lock */
616         reaper = find_child_reaper(father, dead);
617         if (list_empty(&father->children))
618                 return;
619
620         reaper = find_new_reaper(father, reaper);
621         list_for_each_entry(p, &father->children, sibling) {
622                 for_each_thread(p, t) {
623                         RCU_INIT_POINTER(t->real_parent, reaper);
624                         BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
625                         if (likely(!t->ptrace))
626                                 t->parent = t->real_parent;
627                         if (t->pdeath_signal)
628                                 group_send_sig_info(t->pdeath_signal,
629                                                     SEND_SIG_NOINFO, t,
630                                                     PIDTYPE_TGID);
631                 }
632                 /*
633                  * If this is a threaded reparent there is no need to
634                  * notify anyone anything has happened.
635                  */
636                 if (!same_thread_group(reaper, father))
637                         reparent_leader(father, p, dead);
638         }
639         list_splice_tail_init(&father->children, &reaper->children);
640 }
641
642 /*
643  * Send signals to all our closest relatives so that they know
644  * to properly mourn us..
645  */
646 static void exit_notify(struct task_struct *tsk, int group_dead)
647 {
648         bool autoreap;
649         struct task_struct *p, *n;
650         LIST_HEAD(dead);
651
652         write_lock_irq(&tasklist_lock);
653         forget_original_parent(tsk, &dead);
654
655         if (group_dead)
656                 kill_orphaned_pgrp(tsk->group_leader, NULL);
657
658         tsk->exit_state = EXIT_ZOMBIE;
659         if (unlikely(tsk->ptrace)) {
660                 int sig = thread_group_leader(tsk) &&
661                                 thread_group_empty(tsk) &&
662                                 !ptrace_reparented(tsk) ?
663                         tsk->exit_signal : SIGCHLD;
664                 autoreap = do_notify_parent(tsk, sig);
665         } else if (thread_group_leader(tsk)) {
666                 autoreap = thread_group_empty(tsk) &&
667                         do_notify_parent(tsk, tsk->exit_signal);
668         } else {
669                 autoreap = true;
670         }
671
672         if (autoreap) {
673                 tsk->exit_state = EXIT_DEAD;
674                 list_add(&tsk->ptrace_entry, &dead);
675         }
676
677         /* mt-exec, de_thread() is waiting for group leader */
678         if (unlikely(tsk->signal->notify_count < 0))
679                 wake_up_process(tsk->signal->group_exit_task);
680         write_unlock_irq(&tasklist_lock);
681
682         list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
683                 list_del_init(&p->ptrace_entry);
684                 release_task(p);
685         }
686 }
687
688 #ifdef CONFIG_DEBUG_STACK_USAGE
689 static void check_stack_usage(void)
690 {
691         static DEFINE_SPINLOCK(low_water_lock);
692         static int lowest_to_date = THREAD_SIZE;
693         unsigned long free;
694
695         free = stack_not_used(current);
696
697         if (free >= lowest_to_date)
698                 return;
699
700         spin_lock(&low_water_lock);
701         if (free < lowest_to_date) {
702                 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
703                         current->comm, task_pid_nr(current), free);
704                 lowest_to_date = free;
705         }
706         spin_unlock(&low_water_lock);
707 }
708 #else
709 static inline void check_stack_usage(void) {}
710 #endif
711
712 void __noreturn do_exit(long code)
713 {
714         struct task_struct *tsk = current;
715         int group_dead;
716
717         /*
718          * We can get here from a kernel oops, sometimes with preemption off.
719          * Start by checking for critical errors.
720          * Then fix up important state like USER_DS and preemption.
721          * Then do everything else.
722          */
723
724         WARN_ON(blk_needs_flush_plug(tsk));
725
726         if (unlikely(in_interrupt()))
727                 panic("Aiee, killing interrupt handler!");
728         if (unlikely(!tsk->pid))
729                 panic("Attempted to kill the idle task!");
730
731         /*
732          * If do_exit is called because this processes oopsed, it's possible
733          * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
734          * continuing. Amongst other possible reasons, this is to prevent
735          * mm_release()->clear_child_tid() from writing to a user-controlled
736          * kernel address.
737          */
738         force_uaccess_begin();
739
740         if (unlikely(in_atomic())) {
741                 pr_info("note: %s[%d] exited with preempt_count %d\n",
742                         current->comm, task_pid_nr(current),
743                         preempt_count());
744                 preempt_count_set(PREEMPT_ENABLED);
745         }
746
747         profile_task_exit(tsk);
748         kcov_task_exit(tsk);
749
750         ptrace_event(PTRACE_EVENT_EXIT, code);
751
752         validate_creds_for_do_exit(tsk);
753
754         /*
755          * We're taking recursive faults here in do_exit. Safest is to just
756          * leave this task alone and wait for reboot.
757          */
758         if (unlikely(tsk->flags & PF_EXITING)) {
759                 pr_alert("Fixing recursive fault but reboot is needed!\n");
760                 futex_exit_recursive(tsk);
761                 set_current_state(TASK_UNINTERRUPTIBLE);
762                 schedule();
763         }
764
765         exit_signals(tsk);  /* sets PF_EXITING */
766
767         /* sync mm's RSS info before statistics gathering */
768         if (tsk->mm)
769                 sync_mm_rss(tsk->mm);
770         acct_update_integrals(tsk);
771         group_dead = atomic_dec_and_test(&tsk->signal->live);
772         if (group_dead) {
773                 /*
774                  * If the last thread of global init has exited, panic
775                  * immediately to get a useable coredump.
776                  */
777                 if (unlikely(is_global_init(tsk)))
778                         panic("Attempted to kill init! exitcode=0x%08x\n",
779                                 tsk->signal->group_exit_code ?: (int)code);
780
781 #ifdef CONFIG_POSIX_TIMERS
782                 hrtimer_cancel(&tsk->signal->real_timer);
783                 exit_itimers(tsk->signal);
784 #endif
785                 if (tsk->mm)
786                         setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
787         }
788         acct_collect(code, group_dead);
789         if (group_dead)
790                 tty_audit_exit();
791         audit_free(tsk);
792
793         tsk->exit_code = code;
794         taskstats_exit(tsk, group_dead);
795
796         exit_mm();
797
798         if (group_dead)
799                 acct_process();
800         trace_sched_process_exit(tsk);
801
802         exit_sem(tsk);
803         exit_shm(tsk);
804         exit_files(tsk);
805         exit_fs(tsk);
806         if (group_dead)
807                 disassociate_ctty(1);
808         exit_task_namespaces(tsk);
809         exit_task_work(tsk);
810         exit_thread(tsk);
811
812         /*
813          * Flush inherited counters to the parent - before the parent
814          * gets woken up by child-exit notifications.
815          *
816          * because of cgroup mode, must be called before cgroup_exit()
817          */
818         perf_event_exit_task(tsk);
819
820         sched_autogroup_exit_task(tsk);
821         cgroup_exit(tsk);
822
823         /*
824          * FIXME: do that only when needed, using sched_exit tracepoint
825          */
826         flush_ptrace_hw_breakpoint(tsk);
827
828         exit_tasks_rcu_start();
829         exit_notify(tsk, group_dead);
830         proc_exit_connector(tsk);
831         mpol_put_task_policy(tsk);
832 #ifdef CONFIG_FUTEX
833         if (unlikely(current->pi_state_cache))
834                 kfree(current->pi_state_cache);
835 #endif
836         /*
837          * Make sure we are holding no locks:
838          */
839         debug_check_no_locks_held();
840
841         if (tsk->io_context)
842                 exit_io_context(tsk);
843
844         if (tsk->splice_pipe)
845                 free_pipe_info(tsk->splice_pipe);
846
847         if (tsk->task_frag.page)
848                 put_page(tsk->task_frag.page);
849
850         validate_creds_for_do_exit(tsk);
851
852         check_stack_usage();
853         preempt_disable();
854         if (tsk->nr_dirtied)
855                 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
856         exit_rcu();
857         exit_tasks_rcu_finish();
858
859         lockdep_free_task(tsk);
860         do_task_dead();
861 }
862 EXPORT_SYMBOL_GPL(do_exit);
863
864 void complete_and_exit(struct completion *comp, long code)
865 {
866         if (comp)
867                 complete(comp);
868
869         do_exit(code);
870 }
871 EXPORT_SYMBOL(complete_and_exit);
872
873 SYSCALL_DEFINE1(exit, int, error_code)
874 {
875         do_exit((error_code&0xff)<<8);
876 }
877
878 /*
879  * Take down every thread in the group.  This is called by fatal signals
880  * as well as by sys_exit_group (below).
881  */
882 void
883 do_group_exit(int exit_code)
884 {
885         struct signal_struct *sig = current->signal;
886
887         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
888
889         if (signal_group_exit(sig))
890                 exit_code = sig->group_exit_code;
891         else if (!thread_group_empty(current)) {
892                 struct sighand_struct *const sighand = current->sighand;
893
894                 spin_lock_irq(&sighand->siglock);
895                 if (signal_group_exit(sig))
896                         /* Another thread got here before we took the lock.  */
897                         exit_code = sig->group_exit_code;
898                 else {
899                         sig->group_exit_code = exit_code;
900                         sig->flags = SIGNAL_GROUP_EXIT;
901                         zap_other_threads(current);
902                 }
903                 spin_unlock_irq(&sighand->siglock);
904         }
905
906         do_exit(exit_code);
907         /* NOTREACHED */
908 }
909
910 /*
911  * this kills every thread in the thread group. Note that any externally
912  * wait4()-ing process will get the correct exit code - even if this
913  * thread is not the thread group leader.
914  */
915 SYSCALL_DEFINE1(exit_group, int, error_code)
916 {
917         do_group_exit((error_code & 0xff) << 8);
918         /* NOTREACHED */
919         return 0;
920 }
921
922 struct waitid_info {
923         pid_t pid;
924         uid_t uid;
925         int status;
926         int cause;
927 };
928
929 struct wait_opts {
930         enum pid_type           wo_type;
931         int                     wo_flags;
932         struct pid              *wo_pid;
933
934         struct waitid_info      *wo_info;
935         int                     wo_stat;
936         struct rusage           *wo_rusage;
937
938         wait_queue_entry_t              child_wait;
939         int                     notask_error;
940 };
941
942 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
943 {
944         return  wo->wo_type == PIDTYPE_MAX ||
945                 task_pid_type(p, wo->wo_type) == wo->wo_pid;
946 }
947
948 static int
949 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
950 {
951         if (!eligible_pid(wo, p))
952                 return 0;
953
954         /*
955          * Wait for all children (clone and not) if __WALL is set or
956          * if it is traced by us.
957          */
958         if (ptrace || (wo->wo_flags & __WALL))
959                 return 1;
960
961         /*
962          * Otherwise, wait for clone children *only* if __WCLONE is set;
963          * otherwise, wait for non-clone children *only*.
964          *
965          * Note: a "clone" child here is one that reports to its parent
966          * using a signal other than SIGCHLD, or a non-leader thread which
967          * we can only see if it is traced by us.
968          */
969         if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
970                 return 0;
971
972         return 1;
973 }
974
975 /*
976  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
977  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
978  * the lock and this task is uninteresting.  If we return nonzero, we have
979  * released the lock and the system call should return.
980  */
981 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
982 {
983         int state, status;
984         pid_t pid = task_pid_vnr(p);
985         uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
986         struct waitid_info *infop;
987
988         if (!likely(wo->wo_flags & WEXITED))
989                 return 0;
990
991         if (unlikely(wo->wo_flags & WNOWAIT)) {
992                 status = p->exit_code;
993                 get_task_struct(p);
994                 read_unlock(&tasklist_lock);
995                 sched_annotate_sleep();
996                 if (wo->wo_rusage)
997                         getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
998                 put_task_struct(p);
999                 goto out_info;
1000         }
1001         /*
1002          * Move the task's state to DEAD/TRACE, only one thread can do this.
1003          */
1004         state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1005                 EXIT_TRACE : EXIT_DEAD;
1006         if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1007                 return 0;
1008         /*
1009          * We own this thread, nobody else can reap it.
1010          */
1011         read_unlock(&tasklist_lock);
1012         sched_annotate_sleep();
1013
1014         /*
1015          * Check thread_group_leader() to exclude the traced sub-threads.
1016          */
1017         if (state == EXIT_DEAD && thread_group_leader(p)) {
1018                 struct signal_struct *sig = p->signal;
1019                 struct signal_struct *psig = current->signal;
1020                 unsigned long maxrss;
1021                 u64 tgutime, tgstime;
1022
1023                 /*
1024                  * The resource counters for the group leader are in its
1025                  * own task_struct.  Those for dead threads in the group
1026                  * are in its signal_struct, as are those for the child
1027                  * processes it has previously reaped.  All these
1028                  * accumulate in the parent's signal_struct c* fields.
1029                  *
1030                  * We don't bother to take a lock here to protect these
1031                  * p->signal fields because the whole thread group is dead
1032                  * and nobody can change them.
1033                  *
1034                  * psig->stats_lock also protects us from our sub-theads
1035                  * which can reap other children at the same time. Until
1036                  * we change k_getrusage()-like users to rely on this lock
1037                  * we have to take ->siglock as well.
1038                  *
1039                  * We use thread_group_cputime_adjusted() to get times for
1040                  * the thread group, which consolidates times for all threads
1041                  * in the group including the group leader.
1042                  */
1043                 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1044                 spin_lock_irq(&current->sighand->siglock);
1045                 write_seqlock(&psig->stats_lock);
1046                 psig->cutime += tgutime + sig->cutime;
1047                 psig->cstime += tgstime + sig->cstime;
1048                 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1049                 psig->cmin_flt +=
1050                         p->min_flt + sig->min_flt + sig->cmin_flt;
1051                 psig->cmaj_flt +=
1052                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1053                 psig->cnvcsw +=
1054                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1055                 psig->cnivcsw +=
1056                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1057                 psig->cinblock +=
1058                         task_io_get_inblock(p) +
1059                         sig->inblock + sig->cinblock;
1060                 psig->coublock +=
1061                         task_io_get_oublock(p) +
1062                         sig->oublock + sig->coublock;
1063                 maxrss = max(sig->maxrss, sig->cmaxrss);
1064                 if (psig->cmaxrss < maxrss)
1065                         psig->cmaxrss = maxrss;
1066                 task_io_accounting_add(&psig->ioac, &p->ioac);
1067                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1068                 write_sequnlock(&psig->stats_lock);
1069                 spin_unlock_irq(&current->sighand->siglock);
1070         }
1071
1072         if (wo->wo_rusage)
1073                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1074         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1075                 ? p->signal->group_exit_code : p->exit_code;
1076         wo->wo_stat = status;
1077
1078         if (state == EXIT_TRACE) {
1079                 write_lock_irq(&tasklist_lock);
1080                 /* We dropped tasklist, ptracer could die and untrace */
1081                 ptrace_unlink(p);
1082
1083                 /* If parent wants a zombie, don't release it now */
1084                 state = EXIT_ZOMBIE;
1085                 if (do_notify_parent(p, p->exit_signal))
1086                         state = EXIT_DEAD;
1087                 p->exit_state = state;
1088                 write_unlock_irq(&tasklist_lock);
1089         }
1090         if (state == EXIT_DEAD)
1091                 release_task(p);
1092
1093 out_info:
1094         infop = wo->wo_info;
1095         if (infop) {
1096                 if ((status & 0x7f) == 0) {
1097                         infop->cause = CLD_EXITED;
1098                         infop->status = status >> 8;
1099                 } else {
1100                         infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1101                         infop->status = status & 0x7f;
1102                 }
1103                 infop->pid = pid;
1104                 infop->uid = uid;
1105         }
1106
1107         return pid;
1108 }
1109
1110 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1111 {
1112         if (ptrace) {
1113                 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1114                         return &p->exit_code;
1115         } else {
1116                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1117                         return &p->signal->group_exit_code;
1118         }
1119         return NULL;
1120 }
1121
1122 /**
1123  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1124  * @wo: wait options
1125  * @ptrace: is the wait for ptrace
1126  * @p: task to wait for
1127  *
1128  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1129  *
1130  * CONTEXT:
1131  * read_lock(&tasklist_lock), which is released if return value is
1132  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1133  *
1134  * RETURNS:
1135  * 0 if wait condition didn't exist and search for other wait conditions
1136  * should continue.  Non-zero return, -errno on failure and @p's pid on
1137  * success, implies that tasklist_lock is released and wait condition
1138  * search should terminate.
1139  */
1140 static int wait_task_stopped(struct wait_opts *wo,
1141                                 int ptrace, struct task_struct *p)
1142 {
1143         struct waitid_info *infop;
1144         int exit_code, *p_code, why;
1145         uid_t uid = 0; /* unneeded, required by compiler */
1146         pid_t pid;
1147
1148         /*
1149          * Traditionally we see ptrace'd stopped tasks regardless of options.
1150          */
1151         if (!ptrace && !(wo->wo_flags & WUNTRACED))
1152                 return 0;
1153
1154         if (!task_stopped_code(p, ptrace))
1155                 return 0;
1156
1157         exit_code = 0;
1158         spin_lock_irq(&p->sighand->siglock);
1159
1160         p_code = task_stopped_code(p, ptrace);
1161         if (unlikely(!p_code))
1162                 goto unlock_sig;
1163
1164         exit_code = *p_code;
1165         if (!exit_code)
1166                 goto unlock_sig;
1167
1168         if (!unlikely(wo->wo_flags & WNOWAIT))
1169                 *p_code = 0;
1170
1171         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1172 unlock_sig:
1173         spin_unlock_irq(&p->sighand->siglock);
1174         if (!exit_code)
1175                 return 0;
1176
1177         /*
1178          * Now we are pretty sure this task is interesting.
1179          * Make sure it doesn't get reaped out from under us while we
1180          * give up the lock and then examine it below.  We don't want to
1181          * keep holding onto the tasklist_lock while we call getrusage and
1182          * possibly take page faults for user memory.
1183          */
1184         get_task_struct(p);
1185         pid = task_pid_vnr(p);
1186         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1187         read_unlock(&tasklist_lock);
1188         sched_annotate_sleep();
1189         if (wo->wo_rusage)
1190                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1191         put_task_struct(p);
1192
1193         if (likely(!(wo->wo_flags & WNOWAIT)))
1194                 wo->wo_stat = (exit_code << 8) | 0x7f;
1195
1196         infop = wo->wo_info;
1197         if (infop) {
1198                 infop->cause = why;
1199                 infop->status = exit_code;
1200                 infop->pid = pid;
1201                 infop->uid = uid;
1202         }
1203         return pid;
1204 }
1205
1206 /*
1207  * Handle do_wait work for one task in a live, non-stopped state.
1208  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1209  * the lock and this task is uninteresting.  If we return nonzero, we have
1210  * released the lock and the system call should return.
1211  */
1212 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1213 {
1214         struct waitid_info *infop;
1215         pid_t pid;
1216         uid_t uid;
1217
1218         if (!unlikely(wo->wo_flags & WCONTINUED))
1219                 return 0;
1220
1221         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1222                 return 0;
1223
1224         spin_lock_irq(&p->sighand->siglock);
1225         /* Re-check with the lock held.  */
1226         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1227                 spin_unlock_irq(&p->sighand->siglock);
1228                 return 0;
1229         }
1230         if (!unlikely(wo->wo_flags & WNOWAIT))
1231                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1232         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1233         spin_unlock_irq(&p->sighand->siglock);
1234
1235         pid = task_pid_vnr(p);
1236         get_task_struct(p);
1237         read_unlock(&tasklist_lock);
1238         sched_annotate_sleep();
1239         if (wo->wo_rusage)
1240                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1241         put_task_struct(p);
1242
1243         infop = wo->wo_info;
1244         if (!infop) {
1245                 wo->wo_stat = 0xffff;
1246         } else {
1247                 infop->cause = CLD_CONTINUED;
1248                 infop->pid = pid;
1249                 infop->uid = uid;
1250                 infop->status = SIGCONT;
1251         }
1252         return pid;
1253 }
1254
1255 /*
1256  * Consider @p for a wait by @parent.
1257  *
1258  * -ECHILD should be in ->notask_error before the first call.
1259  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1260  * Returns zero if the search for a child should continue;
1261  * then ->notask_error is 0 if @p is an eligible child,
1262  * or still -ECHILD.
1263  */
1264 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1265                                 struct task_struct *p)
1266 {
1267         /*
1268          * We can race with wait_task_zombie() from another thread.
1269          * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1270          * can't confuse the checks below.
1271          */
1272         int exit_state = READ_ONCE(p->exit_state);
1273         int ret;
1274
1275         if (unlikely(exit_state == EXIT_DEAD))
1276                 return 0;
1277
1278         ret = eligible_child(wo, ptrace, p);
1279         if (!ret)
1280                 return ret;
1281
1282         if (unlikely(exit_state == EXIT_TRACE)) {
1283                 /*
1284                  * ptrace == 0 means we are the natural parent. In this case
1285                  * we should clear notask_error, debugger will notify us.
1286                  */
1287                 if (likely(!ptrace))
1288                         wo->notask_error = 0;
1289                 return 0;
1290         }
1291
1292         if (likely(!ptrace) && unlikely(p->ptrace)) {
1293                 /*
1294                  * If it is traced by its real parent's group, just pretend
1295                  * the caller is ptrace_do_wait() and reap this child if it
1296                  * is zombie.
1297                  *
1298                  * This also hides group stop state from real parent; otherwise
1299                  * a single stop can be reported twice as group and ptrace stop.
1300                  * If a ptracer wants to distinguish these two events for its
1301                  * own children it should create a separate process which takes
1302                  * the role of real parent.
1303                  */
1304                 if (!ptrace_reparented(p))
1305                         ptrace = 1;
1306         }
1307
1308         /* slay zombie? */
1309         if (exit_state == EXIT_ZOMBIE) {
1310                 /* we don't reap group leaders with subthreads */
1311                 if (!delay_group_leader(p)) {
1312                         /*
1313                          * A zombie ptracee is only visible to its ptracer.
1314                          * Notification and reaping will be cascaded to the
1315                          * real parent when the ptracer detaches.
1316                          */
1317                         if (unlikely(ptrace) || likely(!p->ptrace))
1318                                 return wait_task_zombie(wo, p);
1319                 }
1320
1321                 /*
1322                  * Allow access to stopped/continued state via zombie by
1323                  * falling through.  Clearing of notask_error is complex.
1324                  *
1325                  * When !@ptrace:
1326                  *
1327                  * If WEXITED is set, notask_error should naturally be
1328                  * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1329                  * so, if there are live subthreads, there are events to
1330                  * wait for.  If all subthreads are dead, it's still safe
1331                  * to clear - this function will be called again in finite
1332                  * amount time once all the subthreads are released and
1333                  * will then return without clearing.
1334                  *
1335                  * When @ptrace:
1336                  *
1337                  * Stopped state is per-task and thus can't change once the
1338                  * target task dies.  Only continued and exited can happen.
1339                  * Clear notask_error if WCONTINUED | WEXITED.
1340                  */
1341                 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1342                         wo->notask_error = 0;
1343         } else {
1344                 /*
1345                  * @p is alive and it's gonna stop, continue or exit, so
1346                  * there always is something to wait for.
1347                  */
1348                 wo->notask_error = 0;
1349         }
1350
1351         /*
1352          * Wait for stopped.  Depending on @ptrace, different stopped state
1353          * is used and the two don't interact with each other.
1354          */
1355         ret = wait_task_stopped(wo, ptrace, p);
1356         if (ret)
1357                 return ret;
1358
1359         /*
1360          * Wait for continued.  There's only one continued state and the
1361          * ptracer can consume it which can confuse the real parent.  Don't
1362          * use WCONTINUED from ptracer.  You don't need or want it.
1363          */
1364         return wait_task_continued(wo, p);
1365 }
1366
1367 /*
1368  * Do the work of do_wait() for one thread in the group, @tsk.
1369  *
1370  * -ECHILD should be in ->notask_error before the first call.
1371  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1372  * Returns zero if the search for a child should continue; then
1373  * ->notask_error is 0 if there were any eligible children,
1374  * or still -ECHILD.
1375  */
1376 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1377 {
1378         struct task_struct *p;
1379
1380         list_for_each_entry(p, &tsk->children, sibling) {
1381                 int ret = wait_consider_task(wo, 0, p);
1382
1383                 if (ret)
1384                         return ret;
1385         }
1386
1387         return 0;
1388 }
1389
1390 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1391 {
1392         struct task_struct *p;
1393
1394         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1395                 int ret = wait_consider_task(wo, 1, p);
1396
1397                 if (ret)
1398                         return ret;
1399         }
1400
1401         return 0;
1402 }
1403
1404 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1405                                 int sync, void *key)
1406 {
1407         struct wait_opts *wo = container_of(wait, struct wait_opts,
1408                                                 child_wait);
1409         struct task_struct *p = key;
1410
1411         if (!eligible_pid(wo, p))
1412                 return 0;
1413
1414         if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1415                 return 0;
1416
1417         return default_wake_function(wait, mode, sync, key);
1418 }
1419
1420 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1421 {
1422         __wake_up_sync_key(&parent->signal->wait_chldexit,
1423                            TASK_INTERRUPTIBLE, p);
1424 }
1425
1426 static long do_wait(struct wait_opts *wo)
1427 {
1428         struct task_struct *tsk;
1429         int retval;
1430
1431         trace_sched_process_wait(wo->wo_pid);
1432
1433         init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1434         wo->child_wait.private = current;
1435         add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1436 repeat:
1437         /*
1438          * If there is nothing that can match our criteria, just get out.
1439          * We will clear ->notask_error to zero if we see any child that
1440          * might later match our criteria, even if we are not able to reap
1441          * it yet.
1442          */
1443         wo->notask_error = -ECHILD;
1444         if ((wo->wo_type < PIDTYPE_MAX) &&
1445            (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1446                 goto notask;
1447
1448         set_current_state(TASK_INTERRUPTIBLE);
1449         read_lock(&tasklist_lock);
1450         tsk = current;
1451         do {
1452                 retval = do_wait_thread(wo, tsk);
1453                 if (retval)
1454                         goto end;
1455
1456                 retval = ptrace_do_wait(wo, tsk);
1457                 if (retval)
1458                         goto end;
1459
1460                 if (wo->wo_flags & __WNOTHREAD)
1461                         break;
1462         } while_each_thread(current, tsk);
1463         read_unlock(&tasklist_lock);
1464
1465 notask:
1466         retval = wo->notask_error;
1467         if (!retval && !(wo->wo_flags & WNOHANG)) {
1468                 retval = -ERESTARTSYS;
1469                 if (!signal_pending(current)) {
1470                         schedule();
1471                         goto repeat;
1472                 }
1473         }
1474 end:
1475         __set_current_state(TASK_RUNNING);
1476         remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1477         return retval;
1478 }
1479
1480 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1481                           int options, struct rusage *ru)
1482 {
1483         struct wait_opts wo;
1484         struct pid *pid = NULL;
1485         enum pid_type type;
1486         long ret;
1487         unsigned int f_flags = 0;
1488
1489         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1490                         __WNOTHREAD|__WCLONE|__WALL))
1491                 return -EINVAL;
1492         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1493                 return -EINVAL;
1494
1495         switch (which) {
1496         case P_ALL:
1497                 type = PIDTYPE_MAX;
1498                 break;
1499         case P_PID:
1500                 type = PIDTYPE_PID;
1501                 if (upid <= 0)
1502                         return -EINVAL;
1503
1504                 pid = find_get_pid(upid);
1505                 break;
1506         case P_PGID:
1507                 type = PIDTYPE_PGID;
1508                 if (upid < 0)
1509                         return -EINVAL;
1510
1511                 if (upid)
1512                         pid = find_get_pid(upid);
1513                 else
1514                         pid = get_task_pid(current, PIDTYPE_PGID);
1515                 break;
1516         case P_PIDFD:
1517                 type = PIDTYPE_PID;
1518                 if (upid < 0)
1519                         return -EINVAL;
1520
1521                 pid = pidfd_get_pid(upid, &f_flags);
1522                 if (IS_ERR(pid))
1523                         return PTR_ERR(pid);
1524
1525                 break;
1526         default:
1527                 return -EINVAL;
1528         }
1529
1530         wo.wo_type      = type;
1531         wo.wo_pid       = pid;
1532         wo.wo_flags     = options;
1533         wo.wo_info      = infop;
1534         wo.wo_rusage    = ru;
1535         if (f_flags & O_NONBLOCK)
1536                 wo.wo_flags |= WNOHANG;
1537
1538         ret = do_wait(&wo);
1539         if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1540                 ret = -EAGAIN;
1541
1542         put_pid(pid);
1543         return ret;
1544 }
1545
1546 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1547                 infop, int, options, struct rusage __user *, ru)
1548 {
1549         struct rusage r;
1550         struct waitid_info info = {.status = 0};
1551         long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1552         int signo = 0;
1553
1554         if (err > 0) {
1555                 signo = SIGCHLD;
1556                 err = 0;
1557                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1558                         return -EFAULT;
1559         }
1560         if (!infop)
1561                 return err;
1562
1563         if (!user_write_access_begin(infop, sizeof(*infop)))
1564                 return -EFAULT;
1565
1566         unsafe_put_user(signo, &infop->si_signo, Efault);
1567         unsafe_put_user(0, &infop->si_errno, Efault);
1568         unsafe_put_user(info.cause, &infop->si_code, Efault);
1569         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1570         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1571         unsafe_put_user(info.status, &infop->si_status, Efault);
1572         user_write_access_end();
1573         return err;
1574 Efault:
1575         user_write_access_end();
1576         return -EFAULT;
1577 }
1578
1579 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1580                   struct rusage *ru)
1581 {
1582         struct wait_opts wo;
1583         struct pid *pid = NULL;
1584         enum pid_type type;
1585         long ret;
1586
1587         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1588                         __WNOTHREAD|__WCLONE|__WALL))
1589                 return -EINVAL;
1590
1591         /* -INT_MIN is not defined */
1592         if (upid == INT_MIN)
1593                 return -ESRCH;
1594
1595         if (upid == -1)
1596                 type = PIDTYPE_MAX;
1597         else if (upid < 0) {
1598                 type = PIDTYPE_PGID;
1599                 pid = find_get_pid(-upid);
1600         } else if (upid == 0) {
1601                 type = PIDTYPE_PGID;
1602                 pid = get_task_pid(current, PIDTYPE_PGID);
1603         } else /* upid > 0 */ {
1604                 type = PIDTYPE_PID;
1605                 pid = find_get_pid(upid);
1606         }
1607
1608         wo.wo_type      = type;
1609         wo.wo_pid       = pid;
1610         wo.wo_flags     = options | WEXITED;
1611         wo.wo_info      = NULL;
1612         wo.wo_stat      = 0;
1613         wo.wo_rusage    = ru;
1614         ret = do_wait(&wo);
1615         put_pid(pid);
1616         if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1617                 ret = -EFAULT;
1618
1619         return ret;
1620 }
1621
1622 int kernel_wait(pid_t pid, int *stat)
1623 {
1624         struct wait_opts wo = {
1625                 .wo_type        = PIDTYPE_PID,
1626                 .wo_pid         = find_get_pid(pid),
1627                 .wo_flags       = WEXITED,
1628         };
1629         int ret;
1630
1631         ret = do_wait(&wo);
1632         if (ret > 0 && wo.wo_stat)
1633                 *stat = wo.wo_stat;
1634         put_pid(wo.wo_pid);
1635         return ret;
1636 }
1637
1638 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1639                 int, options, struct rusage __user *, ru)
1640 {
1641         struct rusage r;
1642         long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1643
1644         if (err > 0) {
1645                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1646                         return -EFAULT;
1647         }
1648         return err;
1649 }
1650
1651 #ifdef __ARCH_WANT_SYS_WAITPID
1652
1653 /*
1654  * sys_waitpid() remains for compatibility. waitpid() should be
1655  * implemented by calling sys_wait4() from libc.a.
1656  */
1657 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1658 {
1659         return kernel_wait4(pid, stat_addr, options, NULL);
1660 }
1661
1662 #endif
1663
1664 #ifdef CONFIG_COMPAT
1665 COMPAT_SYSCALL_DEFINE4(wait4,
1666         compat_pid_t, pid,
1667         compat_uint_t __user *, stat_addr,
1668         int, options,
1669         struct compat_rusage __user *, ru)
1670 {
1671         struct rusage r;
1672         long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1673         if (err > 0) {
1674                 if (ru && put_compat_rusage(&r, ru))
1675                         return -EFAULT;
1676         }
1677         return err;
1678 }
1679
1680 COMPAT_SYSCALL_DEFINE5(waitid,
1681                 int, which, compat_pid_t, pid,
1682                 struct compat_siginfo __user *, infop, int, options,
1683                 struct compat_rusage __user *, uru)
1684 {
1685         struct rusage ru;
1686         struct waitid_info info = {.status = 0};
1687         long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1688         int signo = 0;
1689         if (err > 0) {
1690                 signo = SIGCHLD;
1691                 err = 0;
1692                 if (uru) {
1693                         /* kernel_waitid() overwrites everything in ru */
1694                         if (COMPAT_USE_64BIT_TIME)
1695                                 err = copy_to_user(uru, &ru, sizeof(ru));
1696                         else
1697                                 err = put_compat_rusage(&ru, uru);
1698                         if (err)
1699                                 return -EFAULT;
1700                 }
1701         }
1702
1703         if (!infop)
1704                 return err;
1705
1706         if (!user_write_access_begin(infop, sizeof(*infop)))
1707                 return -EFAULT;
1708
1709         unsafe_put_user(signo, &infop->si_signo, Efault);
1710         unsafe_put_user(0, &infop->si_errno, Efault);
1711         unsafe_put_user(info.cause, &infop->si_code, Efault);
1712         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1713         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1714         unsafe_put_user(info.status, &infop->si_status, Efault);
1715         user_write_access_end();
1716         return err;
1717 Efault:
1718         user_write_access_end();
1719         return -EFAULT;
1720 }
1721 #endif
1722
1723 /**
1724  * thread_group_exited - check that a thread group has exited
1725  * @pid: tgid of thread group to be checked.
1726  *
1727  * Test if the thread group represented by tgid has exited (all
1728  * threads are zombies, dead or completely gone).
1729  *
1730  * Return: true if the thread group has exited. false otherwise.
1731  */
1732 bool thread_group_exited(struct pid *pid)
1733 {
1734         struct task_struct *task;
1735         bool exited;
1736
1737         rcu_read_lock();
1738         task = pid_task(pid, PIDTYPE_PID);
1739         exited = !task ||
1740                 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1741         rcu_read_unlock();
1742
1743         return exited;
1744 }
1745 EXPORT_SYMBOL(thread_group_exited);
1746
1747 __weak void abort(void)
1748 {
1749         BUG();
1750
1751         /* if that doesn't kill us, halt */
1752         panic("Oops failed to kill thread");
1753 }
1754 EXPORT_SYMBOL(abort);