futex: Set task::futex_state to DEAD right after handling futex exit
[platform/kernel/linux-rpi.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/pgtable.h>
70 #include <asm/mmu_context.h>
71
72 static void __unhash_process(struct task_struct *p, bool group_dead)
73 {
74         nr_threads--;
75         detach_pid(p, PIDTYPE_PID);
76         if (group_dead) {
77                 detach_pid(p, PIDTYPE_TGID);
78                 detach_pid(p, PIDTYPE_PGID);
79                 detach_pid(p, PIDTYPE_SID);
80
81                 list_del_rcu(&p->tasks);
82                 list_del_init(&p->sibling);
83                 __this_cpu_dec(process_counts);
84         }
85         list_del_rcu(&p->thread_group);
86         list_del_rcu(&p->thread_node);
87 }
88
89 /*
90  * This function expects the tasklist_lock write-locked.
91  */
92 static void __exit_signal(struct task_struct *tsk)
93 {
94         struct signal_struct *sig = tsk->signal;
95         bool group_dead = thread_group_leader(tsk);
96         struct sighand_struct *sighand;
97         struct tty_struct *uninitialized_var(tty);
98         u64 utime, stime;
99
100         sighand = rcu_dereference_check(tsk->sighand,
101                                         lockdep_tasklist_lock_is_held());
102         spin_lock(&sighand->siglock);
103
104 #ifdef CONFIG_POSIX_TIMERS
105         posix_cpu_timers_exit(tsk);
106         if (group_dead) {
107                 posix_cpu_timers_exit_group(tsk);
108         } else {
109                 /*
110                  * This can only happen if the caller is de_thread().
111                  * FIXME: this is the temporary hack, we should teach
112                  * posix-cpu-timers to handle this case correctly.
113                  */
114                 if (unlikely(has_group_leader_pid(tsk)))
115                         posix_cpu_timers_exit_group(tsk);
116         }
117 #endif
118
119         if (group_dead) {
120                 tty = sig->tty;
121                 sig->tty = NULL;
122         } else {
123                 /*
124                  * If there is any task waiting for the group exit
125                  * then notify it:
126                  */
127                 if (sig->notify_count > 0 && !--sig->notify_count)
128                         wake_up_process(sig->group_exit_task);
129
130                 if (tsk == sig->curr_target)
131                         sig->curr_target = next_thread(tsk);
132         }
133
134         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
135                               sizeof(unsigned long long));
136
137         /*
138          * Accumulate here the counters for all threads as they die. We could
139          * skip the group leader because it is the last user of signal_struct,
140          * but we want to avoid the race with thread_group_cputime() which can
141          * see the empty ->thread_head list.
142          */
143         task_cputime(tsk, &utime, &stime);
144         write_seqlock(&sig->stats_lock);
145         sig->utime += utime;
146         sig->stime += stime;
147         sig->gtime += task_gtime(tsk);
148         sig->min_flt += tsk->min_flt;
149         sig->maj_flt += tsk->maj_flt;
150         sig->nvcsw += tsk->nvcsw;
151         sig->nivcsw += tsk->nivcsw;
152         sig->inblock += task_io_get_inblock(tsk);
153         sig->oublock += task_io_get_oublock(tsk);
154         task_io_accounting_add(&sig->ioac, &tsk->ioac);
155         sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
156         sig->nr_threads--;
157         __unhash_process(tsk, group_dead);
158         write_sequnlock(&sig->stats_lock);
159
160         /*
161          * Do this under ->siglock, we can race with another thread
162          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
163          */
164         flush_sigqueue(&tsk->pending);
165         tsk->sighand = NULL;
166         spin_unlock(&sighand->siglock);
167
168         __cleanup_sighand(sighand);
169         clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
170         if (group_dead) {
171                 flush_sigqueue(&sig->shared_pending);
172                 tty_kref_put(tty);
173         }
174 }
175
176 static void delayed_put_task_struct(struct rcu_head *rhp)
177 {
178         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
179
180         perf_event_delayed_put(tsk);
181         trace_sched_process_free(tsk);
182         put_task_struct(tsk);
183 }
184
185 void put_task_struct_rcu_user(struct task_struct *task)
186 {
187         if (refcount_dec_and_test(&task->rcu_users))
188                 call_rcu(&task->rcu, delayed_put_task_struct);
189 }
190
191 void release_task(struct task_struct *p)
192 {
193         struct task_struct *leader;
194         int zap_leader;
195 repeat:
196         /* don't need to get the RCU readlock here - the process is dead and
197          * can't be modifying its own credentials. But shut RCU-lockdep up */
198         rcu_read_lock();
199         atomic_dec(&__task_cred(p)->user->processes);
200         rcu_read_unlock();
201
202         proc_flush_task(p);
203         cgroup_release(p);
204
205         write_lock_irq(&tasklist_lock);
206         ptrace_release_task(p);
207         __exit_signal(p);
208
209         /*
210          * If we are the last non-leader member of the thread
211          * group, and the leader is zombie, then notify the
212          * group leader's parent process. (if it wants notification.)
213          */
214         zap_leader = 0;
215         leader = p->group_leader;
216         if (leader != p && thread_group_empty(leader)
217                         && leader->exit_state == EXIT_ZOMBIE) {
218                 /*
219                  * If we were the last child thread and the leader has
220                  * exited already, and the leader's parent ignores SIGCHLD,
221                  * then we are the one who should release the leader.
222                  */
223                 zap_leader = do_notify_parent(leader, leader->exit_signal);
224                 if (zap_leader)
225                         leader->exit_state = EXIT_DEAD;
226         }
227
228         write_unlock_irq(&tasklist_lock);
229         release_thread(p);
230         put_task_struct_rcu_user(p);
231
232         p = leader;
233         if (unlikely(zap_leader))
234                 goto repeat;
235 }
236
237 void rcuwait_wake_up(struct rcuwait *w)
238 {
239         struct task_struct *task;
240
241         rcu_read_lock();
242
243         /*
244          * Order condition vs @task, such that everything prior to the load
245          * of @task is visible. This is the condition as to why the user called
246          * rcuwait_trywake() in the first place. Pairs with set_current_state()
247          * barrier (A) in rcuwait_wait_event().
248          *
249          *    WAIT                WAKE
250          *    [S] tsk = current   [S] cond = true
251          *        MB (A)              MB (B)
252          *    [L] cond            [L] tsk
253          */
254         smp_mb(); /* (B) */
255
256         task = rcu_dereference(w->task);
257         if (task)
258                 wake_up_process(task);
259         rcu_read_unlock();
260 }
261
262 /*
263  * Determine if a process group is "orphaned", according to the POSIX
264  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
265  * by terminal-generated stop signals.  Newly orphaned process groups are
266  * to receive a SIGHUP and a SIGCONT.
267  *
268  * "I ask you, have you ever known what it is to be an orphan?"
269  */
270 static int will_become_orphaned_pgrp(struct pid *pgrp,
271                                         struct task_struct *ignored_task)
272 {
273         struct task_struct *p;
274
275         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
276                 if ((p == ignored_task) ||
277                     (p->exit_state && thread_group_empty(p)) ||
278                     is_global_init(p->real_parent))
279                         continue;
280
281                 if (task_pgrp(p->real_parent) != pgrp &&
282                     task_session(p->real_parent) == task_session(p))
283                         return 0;
284         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
285
286         return 1;
287 }
288
289 int is_current_pgrp_orphaned(void)
290 {
291         int retval;
292
293         read_lock(&tasklist_lock);
294         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
295         read_unlock(&tasklist_lock);
296
297         return retval;
298 }
299
300 static bool has_stopped_jobs(struct pid *pgrp)
301 {
302         struct task_struct *p;
303
304         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
305                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
306                         return true;
307         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
308
309         return false;
310 }
311
312 /*
313  * Check to see if any process groups have become orphaned as
314  * a result of our exiting, and if they have any stopped jobs,
315  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
316  */
317 static void
318 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
319 {
320         struct pid *pgrp = task_pgrp(tsk);
321         struct task_struct *ignored_task = tsk;
322
323         if (!parent)
324                 /* exit: our father is in a different pgrp than
325                  * we are and we were the only connection outside.
326                  */
327                 parent = tsk->real_parent;
328         else
329                 /* reparent: our child is in a different pgrp than
330                  * we are, and it was the only connection outside.
331                  */
332                 ignored_task = NULL;
333
334         if (task_pgrp(parent) != pgrp &&
335             task_session(parent) == task_session(tsk) &&
336             will_become_orphaned_pgrp(pgrp, ignored_task) &&
337             has_stopped_jobs(pgrp)) {
338                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
339                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
340         }
341 }
342
343 #ifdef CONFIG_MEMCG
344 /*
345  * A task is exiting.   If it owned this mm, find a new owner for the mm.
346  */
347 void mm_update_next_owner(struct mm_struct *mm)
348 {
349         struct task_struct *c, *g, *p = current;
350
351 retry:
352         /*
353          * If the exiting or execing task is not the owner, it's
354          * someone else's problem.
355          */
356         if (mm->owner != p)
357                 return;
358         /*
359          * The current owner is exiting/execing and there are no other
360          * candidates.  Do not leave the mm pointing to a possibly
361          * freed task structure.
362          */
363         if (atomic_read(&mm->mm_users) <= 1) {
364                 WRITE_ONCE(mm->owner, NULL);
365                 return;
366         }
367
368         read_lock(&tasklist_lock);
369         /*
370          * Search in the children
371          */
372         list_for_each_entry(c, &p->children, sibling) {
373                 if (c->mm == mm)
374                         goto assign_new_owner;
375         }
376
377         /*
378          * Search in the siblings
379          */
380         list_for_each_entry(c, &p->real_parent->children, sibling) {
381                 if (c->mm == mm)
382                         goto assign_new_owner;
383         }
384
385         /*
386          * Search through everything else, we should not get here often.
387          */
388         for_each_process(g) {
389                 if (g->flags & PF_KTHREAD)
390                         continue;
391                 for_each_thread(g, c) {
392                         if (c->mm == mm)
393                                 goto assign_new_owner;
394                         if (c->mm)
395                                 break;
396                 }
397         }
398         read_unlock(&tasklist_lock);
399         /*
400          * We found no owner yet mm_users > 1: this implies that we are
401          * most likely racing with swapoff (try_to_unuse()) or /proc or
402          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
403          */
404         WRITE_ONCE(mm->owner, NULL);
405         return;
406
407 assign_new_owner:
408         BUG_ON(c == p);
409         get_task_struct(c);
410         /*
411          * The task_lock protects c->mm from changing.
412          * We always want mm->owner->mm == mm
413          */
414         task_lock(c);
415         /*
416          * Delay read_unlock() till we have the task_lock()
417          * to ensure that c does not slip away underneath us
418          */
419         read_unlock(&tasklist_lock);
420         if (c->mm != mm) {
421                 task_unlock(c);
422                 put_task_struct(c);
423                 goto retry;
424         }
425         WRITE_ONCE(mm->owner, c);
426         task_unlock(c);
427         put_task_struct(c);
428 }
429 #endif /* CONFIG_MEMCG */
430
431 /*
432  * Turn us into a lazy TLB process if we
433  * aren't already..
434  */
435 static void exit_mm(void)
436 {
437         struct mm_struct *mm = current->mm;
438         struct core_state *core_state;
439
440         exit_mm_release(current, mm);
441         if (!mm)
442                 return;
443         sync_mm_rss(mm);
444         /*
445          * Serialize with any possible pending coredump.
446          * We must hold mmap_sem around checking core_state
447          * and clearing tsk->mm.  The core-inducing thread
448          * will increment ->nr_threads for each thread in the
449          * group with ->mm != NULL.
450          */
451         down_read(&mm->mmap_sem);
452         core_state = mm->core_state;
453         if (core_state) {
454                 struct core_thread self;
455
456                 up_read(&mm->mmap_sem);
457
458                 self.task = current;
459                 self.next = xchg(&core_state->dumper.next, &self);
460                 /*
461                  * Implies mb(), the result of xchg() must be visible
462                  * to core_state->dumper.
463                  */
464                 if (atomic_dec_and_test(&core_state->nr_threads))
465                         complete(&core_state->startup);
466
467                 for (;;) {
468                         set_current_state(TASK_UNINTERRUPTIBLE);
469                         if (!self.task) /* see coredump_finish() */
470                                 break;
471                         freezable_schedule();
472                 }
473                 __set_current_state(TASK_RUNNING);
474                 down_read(&mm->mmap_sem);
475         }
476         mmgrab(mm);
477         BUG_ON(mm != current->active_mm);
478         /* more a memory barrier than a real lock */
479         task_lock(current);
480         current->mm = NULL;
481         up_read(&mm->mmap_sem);
482         enter_lazy_tlb(mm, current);
483         task_unlock(current);
484         mm_update_next_owner(mm);
485         mmput(mm);
486         if (test_thread_flag(TIF_MEMDIE))
487                 exit_oom_victim();
488 }
489
490 static struct task_struct *find_alive_thread(struct task_struct *p)
491 {
492         struct task_struct *t;
493
494         for_each_thread(p, t) {
495                 if (!(t->flags & PF_EXITING))
496                         return t;
497         }
498         return NULL;
499 }
500
501 static struct task_struct *find_child_reaper(struct task_struct *father,
502                                                 struct list_head *dead)
503         __releases(&tasklist_lock)
504         __acquires(&tasklist_lock)
505 {
506         struct pid_namespace *pid_ns = task_active_pid_ns(father);
507         struct task_struct *reaper = pid_ns->child_reaper;
508         struct task_struct *p, *n;
509
510         if (likely(reaper != father))
511                 return reaper;
512
513         reaper = find_alive_thread(father);
514         if (reaper) {
515                 pid_ns->child_reaper = reaper;
516                 return reaper;
517         }
518
519         write_unlock_irq(&tasklist_lock);
520         if (unlikely(pid_ns == &init_pid_ns)) {
521                 panic("Attempted to kill init! exitcode=0x%08x\n",
522                         father->signal->group_exit_code ?: father->exit_code);
523         }
524
525         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
526                 list_del_init(&p->ptrace_entry);
527                 release_task(p);
528         }
529
530         zap_pid_ns_processes(pid_ns);
531         write_lock_irq(&tasklist_lock);
532
533         return father;
534 }
535
536 /*
537  * When we die, we re-parent all our children, and try to:
538  * 1. give them to another thread in our thread group, if such a member exists
539  * 2. give it to the first ancestor process which prctl'd itself as a
540  *    child_subreaper for its children (like a service manager)
541  * 3. give it to the init process (PID 1) in our pid namespace
542  */
543 static struct task_struct *find_new_reaper(struct task_struct *father,
544                                            struct task_struct *child_reaper)
545 {
546         struct task_struct *thread, *reaper;
547
548         thread = find_alive_thread(father);
549         if (thread)
550                 return thread;
551
552         if (father->signal->has_child_subreaper) {
553                 unsigned int ns_level = task_pid(father)->level;
554                 /*
555                  * Find the first ->is_child_subreaper ancestor in our pid_ns.
556                  * We can't check reaper != child_reaper to ensure we do not
557                  * cross the namespaces, the exiting parent could be injected
558                  * by setns() + fork().
559                  * We check pid->level, this is slightly more efficient than
560                  * task_active_pid_ns(reaper) != task_active_pid_ns(father).
561                  */
562                 for (reaper = father->real_parent;
563                      task_pid(reaper)->level == ns_level;
564                      reaper = reaper->real_parent) {
565                         if (reaper == &init_task)
566                                 break;
567                         if (!reaper->signal->is_child_subreaper)
568                                 continue;
569                         thread = find_alive_thread(reaper);
570                         if (thread)
571                                 return thread;
572                 }
573         }
574
575         return child_reaper;
576 }
577
578 /*
579 * Any that need to be release_task'd are put on the @dead list.
580  */
581 static void reparent_leader(struct task_struct *father, struct task_struct *p,
582                                 struct list_head *dead)
583 {
584         if (unlikely(p->exit_state == EXIT_DEAD))
585                 return;
586
587         /* We don't want people slaying init. */
588         p->exit_signal = SIGCHLD;
589
590         /* If it has exited notify the new parent about this child's death. */
591         if (!p->ptrace &&
592             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
593                 if (do_notify_parent(p, p->exit_signal)) {
594                         p->exit_state = EXIT_DEAD;
595                         list_add(&p->ptrace_entry, dead);
596                 }
597         }
598
599         kill_orphaned_pgrp(p, father);
600 }
601
602 /*
603  * This does two things:
604  *
605  * A.  Make init inherit all the child processes
606  * B.  Check to see if any process groups have become orphaned
607  *      as a result of our exiting, and if they have any stopped
608  *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
609  */
610 static void forget_original_parent(struct task_struct *father,
611                                         struct list_head *dead)
612 {
613         struct task_struct *p, *t, *reaper;
614
615         if (unlikely(!list_empty(&father->ptraced)))
616                 exit_ptrace(father, dead);
617
618         /* Can drop and reacquire tasklist_lock */
619         reaper = find_child_reaper(father, dead);
620         if (list_empty(&father->children))
621                 return;
622
623         reaper = find_new_reaper(father, reaper);
624         list_for_each_entry(p, &father->children, sibling) {
625                 for_each_thread(p, t) {
626                         t->real_parent = reaper;
627                         BUG_ON((!t->ptrace) != (t->parent == father));
628                         if (likely(!t->ptrace))
629                                 t->parent = t->real_parent;
630                         if (t->pdeath_signal)
631                                 group_send_sig_info(t->pdeath_signal,
632                                                     SEND_SIG_NOINFO, t,
633                                                     PIDTYPE_TGID);
634                 }
635                 /*
636                  * If this is a threaded reparent there is no need to
637                  * notify anyone anything has happened.
638                  */
639                 if (!same_thread_group(reaper, father))
640                         reparent_leader(father, p, dead);
641         }
642         list_splice_tail_init(&father->children, &reaper->children);
643 }
644
645 /*
646  * Send signals to all our closest relatives so that they know
647  * to properly mourn us..
648  */
649 static void exit_notify(struct task_struct *tsk, int group_dead)
650 {
651         bool autoreap;
652         struct task_struct *p, *n;
653         LIST_HEAD(dead);
654
655         write_lock_irq(&tasklist_lock);
656         forget_original_parent(tsk, &dead);
657
658         if (group_dead)
659                 kill_orphaned_pgrp(tsk->group_leader, NULL);
660
661         tsk->exit_state = EXIT_ZOMBIE;
662         if (unlikely(tsk->ptrace)) {
663                 int sig = thread_group_leader(tsk) &&
664                                 thread_group_empty(tsk) &&
665                                 !ptrace_reparented(tsk) ?
666                         tsk->exit_signal : SIGCHLD;
667                 autoreap = do_notify_parent(tsk, sig);
668         } else if (thread_group_leader(tsk)) {
669                 autoreap = thread_group_empty(tsk) &&
670                         do_notify_parent(tsk, tsk->exit_signal);
671         } else {
672                 autoreap = true;
673         }
674
675         if (autoreap) {
676                 tsk->exit_state = EXIT_DEAD;
677                 list_add(&tsk->ptrace_entry, &dead);
678         }
679
680         /* mt-exec, de_thread() is waiting for group leader */
681         if (unlikely(tsk->signal->notify_count < 0))
682                 wake_up_process(tsk->signal->group_exit_task);
683         write_unlock_irq(&tasklist_lock);
684
685         list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
686                 list_del_init(&p->ptrace_entry);
687                 release_task(p);
688         }
689 }
690
691 #ifdef CONFIG_DEBUG_STACK_USAGE
692 static void check_stack_usage(void)
693 {
694         static DEFINE_SPINLOCK(low_water_lock);
695         static int lowest_to_date = THREAD_SIZE;
696         unsigned long free;
697
698         free = stack_not_used(current);
699
700         if (free >= lowest_to_date)
701                 return;
702
703         spin_lock(&low_water_lock);
704         if (free < lowest_to_date) {
705                 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
706                         current->comm, task_pid_nr(current), free);
707                 lowest_to_date = free;
708         }
709         spin_unlock(&low_water_lock);
710 }
711 #else
712 static inline void check_stack_usage(void) {}
713 #endif
714
715 void __noreturn do_exit(long code)
716 {
717         struct task_struct *tsk = current;
718         int group_dead;
719
720         profile_task_exit(tsk);
721         kcov_task_exit(tsk);
722
723         WARN_ON(blk_needs_flush_plug(tsk));
724
725         if (unlikely(in_interrupt()))
726                 panic("Aiee, killing interrupt handler!");
727         if (unlikely(!tsk->pid))
728                 panic("Attempted to kill the idle task!");
729
730         /*
731          * If do_exit is called because this processes oopsed, it's possible
732          * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
733          * continuing. Amongst other possible reasons, this is to prevent
734          * mm_release()->clear_child_tid() from writing to a user-controlled
735          * kernel address.
736          */
737         set_fs(USER_DS);
738
739         ptrace_event(PTRACE_EVENT_EXIT, code);
740
741         validate_creds_for_do_exit(tsk);
742
743         /*
744          * We're taking recursive faults here in do_exit. Safest is to just
745          * leave this task alone and wait for reboot.
746          */
747         if (unlikely(tsk->flags & PF_EXITING)) {
748                 pr_alert("Fixing recursive fault but reboot is needed!\n");
749                 futex_exit_done(tsk);
750                 set_current_state(TASK_UNINTERRUPTIBLE);
751                 schedule();
752         }
753
754         exit_signals(tsk);  /* sets PF_EXITING */
755         /*
756          * Ensure that all new tsk->pi_lock acquisitions must observe
757          * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
758          */
759         smp_mb();
760         /*
761          * Ensure that we must observe the pi_state in exit_mm() ->
762          * mm_release() -> exit_pi_state_list().
763          */
764         raw_spin_lock_irq(&tsk->pi_lock);
765         raw_spin_unlock_irq(&tsk->pi_lock);
766
767         if (unlikely(in_atomic())) {
768                 pr_info("note: %s[%d] exited with preempt_count %d\n",
769                         current->comm, task_pid_nr(current),
770                         preempt_count());
771                 preempt_count_set(PREEMPT_ENABLED);
772         }
773
774         /* sync mm's RSS info before statistics gathering */
775         if (tsk->mm)
776                 sync_mm_rss(tsk->mm);
777         acct_update_integrals(tsk);
778         group_dead = atomic_dec_and_test(&tsk->signal->live);
779         if (group_dead) {
780 #ifdef CONFIG_POSIX_TIMERS
781                 hrtimer_cancel(&tsk->signal->real_timer);
782                 exit_itimers(tsk->signal);
783 #endif
784                 if (tsk->mm)
785                         setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
786         }
787         acct_collect(code, group_dead);
788         if (group_dead)
789                 tty_audit_exit();
790         audit_free(tsk);
791
792         tsk->exit_code = code;
793         taskstats_exit(tsk, group_dead);
794
795         exit_mm();
796
797         if (group_dead)
798                 acct_process();
799         trace_sched_process_exit(tsk);
800
801         exit_sem(tsk);
802         exit_shm(tsk);
803         exit_files(tsk);
804         exit_fs(tsk);
805         if (group_dead)
806                 disassociate_ctty(1);
807         exit_task_namespaces(tsk);
808         exit_task_work(tsk);
809         exit_thread(tsk);
810         exit_umh(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, 1, 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 || hlist_empty(&wo->wo_pid->tasks[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 struct pid *pidfd_get_pid(unsigned int fd)
1481 {
1482         struct fd f;
1483         struct pid *pid;
1484
1485         f = fdget(fd);
1486         if (!f.file)
1487                 return ERR_PTR(-EBADF);
1488
1489         pid = pidfd_pid(f.file);
1490         if (!IS_ERR(pid))
1491                 get_pid(pid);
1492
1493         fdput(f);
1494         return pid;
1495 }
1496
1497 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1498                           int options, struct rusage *ru)
1499 {
1500         struct wait_opts wo;
1501         struct pid *pid = NULL;
1502         enum pid_type type;
1503         long ret;
1504
1505         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1506                         __WNOTHREAD|__WCLONE|__WALL))
1507                 return -EINVAL;
1508         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1509                 return -EINVAL;
1510
1511         switch (which) {
1512         case P_ALL:
1513                 type = PIDTYPE_MAX;
1514                 break;
1515         case P_PID:
1516                 type = PIDTYPE_PID;
1517                 if (upid <= 0)
1518                         return -EINVAL;
1519
1520                 pid = find_get_pid(upid);
1521                 break;
1522         case P_PGID:
1523                 type = PIDTYPE_PGID;
1524                 if (upid < 0)
1525                         return -EINVAL;
1526
1527                 if (upid)
1528                         pid = find_get_pid(upid);
1529                 else
1530                         pid = get_task_pid(current, PIDTYPE_PGID);
1531                 break;
1532         case P_PIDFD:
1533                 type = PIDTYPE_PID;
1534                 if (upid < 0)
1535                         return -EINVAL;
1536
1537                 pid = pidfd_get_pid(upid);
1538                 if (IS_ERR(pid))
1539                         return PTR_ERR(pid);
1540                 break;
1541         default:
1542                 return -EINVAL;
1543         }
1544
1545         wo.wo_type      = type;
1546         wo.wo_pid       = pid;
1547         wo.wo_flags     = options;
1548         wo.wo_info      = infop;
1549         wo.wo_rusage    = ru;
1550         ret = do_wait(&wo);
1551
1552         put_pid(pid);
1553         return ret;
1554 }
1555
1556 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1557                 infop, int, options, struct rusage __user *, ru)
1558 {
1559         struct rusage r;
1560         struct waitid_info info = {.status = 0};
1561         long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1562         int signo = 0;
1563
1564         if (err > 0) {
1565                 signo = SIGCHLD;
1566                 err = 0;
1567                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1568                         return -EFAULT;
1569         }
1570         if (!infop)
1571                 return err;
1572
1573         if (!user_access_begin(infop, sizeof(*infop)))
1574                 return -EFAULT;
1575
1576         unsafe_put_user(signo, &infop->si_signo, Efault);
1577         unsafe_put_user(0, &infop->si_errno, Efault);
1578         unsafe_put_user(info.cause, &infop->si_code, Efault);
1579         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1580         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1581         unsafe_put_user(info.status, &infop->si_status, Efault);
1582         user_access_end();
1583         return err;
1584 Efault:
1585         user_access_end();
1586         return -EFAULT;
1587 }
1588
1589 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1590                   struct rusage *ru)
1591 {
1592         struct wait_opts wo;
1593         struct pid *pid = NULL;
1594         enum pid_type type;
1595         long ret;
1596
1597         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1598                         __WNOTHREAD|__WCLONE|__WALL))
1599                 return -EINVAL;
1600
1601         /* -INT_MIN is not defined */
1602         if (upid == INT_MIN)
1603                 return -ESRCH;
1604
1605         if (upid == -1)
1606                 type = PIDTYPE_MAX;
1607         else if (upid < 0) {
1608                 type = PIDTYPE_PGID;
1609                 pid = find_get_pid(-upid);
1610         } else if (upid == 0) {
1611                 type = PIDTYPE_PGID;
1612                 pid = get_task_pid(current, PIDTYPE_PGID);
1613         } else /* upid > 0 */ {
1614                 type = PIDTYPE_PID;
1615                 pid = find_get_pid(upid);
1616         }
1617
1618         wo.wo_type      = type;
1619         wo.wo_pid       = pid;
1620         wo.wo_flags     = options | WEXITED;
1621         wo.wo_info      = NULL;
1622         wo.wo_stat      = 0;
1623         wo.wo_rusage    = ru;
1624         ret = do_wait(&wo);
1625         put_pid(pid);
1626         if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1627                 ret = -EFAULT;
1628
1629         return ret;
1630 }
1631
1632 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1633                 int, options, struct rusage __user *, ru)
1634 {
1635         struct rusage r;
1636         long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1637
1638         if (err > 0) {
1639                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1640                         return -EFAULT;
1641         }
1642         return err;
1643 }
1644
1645 #ifdef __ARCH_WANT_SYS_WAITPID
1646
1647 /*
1648  * sys_waitpid() remains for compatibility. waitpid() should be
1649  * implemented by calling sys_wait4() from libc.a.
1650  */
1651 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1652 {
1653         return kernel_wait4(pid, stat_addr, options, NULL);
1654 }
1655
1656 #endif
1657
1658 #ifdef CONFIG_COMPAT
1659 COMPAT_SYSCALL_DEFINE4(wait4,
1660         compat_pid_t, pid,
1661         compat_uint_t __user *, stat_addr,
1662         int, options,
1663         struct compat_rusage __user *, ru)
1664 {
1665         struct rusage r;
1666         long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1667         if (err > 0) {
1668                 if (ru && put_compat_rusage(&r, ru))
1669                         return -EFAULT;
1670         }
1671         return err;
1672 }
1673
1674 COMPAT_SYSCALL_DEFINE5(waitid,
1675                 int, which, compat_pid_t, pid,
1676                 struct compat_siginfo __user *, infop, int, options,
1677                 struct compat_rusage __user *, uru)
1678 {
1679         struct rusage ru;
1680         struct waitid_info info = {.status = 0};
1681         long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1682         int signo = 0;
1683         if (err > 0) {
1684                 signo = SIGCHLD;
1685                 err = 0;
1686                 if (uru) {
1687                         /* kernel_waitid() overwrites everything in ru */
1688                         if (COMPAT_USE_64BIT_TIME)
1689                                 err = copy_to_user(uru, &ru, sizeof(ru));
1690                         else
1691                                 err = put_compat_rusage(&ru, uru);
1692                         if (err)
1693                                 return -EFAULT;
1694                 }
1695         }
1696
1697         if (!infop)
1698                 return err;
1699
1700         if (!user_access_begin(infop, sizeof(*infop)))
1701                 return -EFAULT;
1702
1703         unsafe_put_user(signo, &infop->si_signo, Efault);
1704         unsafe_put_user(0, &infop->si_errno, Efault);
1705         unsafe_put_user(info.cause, &infop->si_code, Efault);
1706         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1707         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1708         unsafe_put_user(info.status, &infop->si_status, Efault);
1709         user_access_end();
1710         return err;
1711 Efault:
1712         user_access_end();
1713         return -EFAULT;
1714 }
1715 #endif
1716
1717 __weak void abort(void)
1718 {
1719         BUG();
1720
1721         /* if that doesn't kill us, halt */
1722         panic("Oops failed to kill thread");
1723 }
1724 EXPORT_SYMBOL(abort);