x86/ioapic: Cleanup the timer_works() irqflags mess
[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                 self.next = xchg(&core_state->dumper.next, &self);
458                 /*
459                  * Implies mb(), the result of xchg() must be visible
460                  * to core_state->dumper.
461                  */
462                 if (atomic_dec_and_test(&core_state->nr_threads))
463                         complete(&core_state->startup);
464
465                 for (;;) {
466                         set_current_state(TASK_UNINTERRUPTIBLE);
467                         if (!self.task) /* see coredump_finish() */
468                                 break;
469                         freezable_schedule();
470                 }
471                 __set_current_state(TASK_RUNNING);
472                 mmap_read_lock(mm);
473         }
474         mmgrab(mm);
475         BUG_ON(mm != current->active_mm);
476         /* more a memory barrier than a real lock */
477         task_lock(current);
478         current->mm = NULL;
479         mmap_read_unlock(mm);
480         enter_lazy_tlb(mm, current);
481         task_unlock(current);
482         mm_update_next_owner(mm);
483         mmput(mm);
484         if (test_thread_flag(TIF_MEMDIE))
485                 exit_oom_victim();
486 }
487
488 static struct task_struct *find_alive_thread(struct task_struct *p)
489 {
490         struct task_struct *t;
491
492         for_each_thread(p, t) {
493                 if (!(t->flags & PF_EXITING))
494                         return t;
495         }
496         return NULL;
497 }
498
499 static struct task_struct *find_child_reaper(struct task_struct *father,
500                                                 struct list_head *dead)
501         __releases(&tasklist_lock)
502         __acquires(&tasklist_lock)
503 {
504         struct pid_namespace *pid_ns = task_active_pid_ns(father);
505         struct task_struct *reaper = pid_ns->child_reaper;
506         struct task_struct *p, *n;
507
508         if (likely(reaper != father))
509                 return reaper;
510
511         reaper = find_alive_thread(father);
512         if (reaper) {
513                 pid_ns->child_reaper = reaper;
514                 return reaper;
515         }
516
517         write_unlock_irq(&tasklist_lock);
518
519         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
520                 list_del_init(&p->ptrace_entry);
521                 release_task(p);
522         }
523
524         zap_pid_ns_processes(pid_ns);
525         write_lock_irq(&tasklist_lock);
526
527         return father;
528 }
529
530 /*
531  * When we die, we re-parent all our children, and try to:
532  * 1. give them to another thread in our thread group, if such a member exists
533  * 2. give it to the first ancestor process which prctl'd itself as a
534  *    child_subreaper for its children (like a service manager)
535  * 3. give it to the init process (PID 1) in our pid namespace
536  */
537 static struct task_struct *find_new_reaper(struct task_struct *father,
538                                            struct task_struct *child_reaper)
539 {
540         struct task_struct *thread, *reaper;
541
542         thread = find_alive_thread(father);
543         if (thread)
544                 return thread;
545
546         if (father->signal->has_child_subreaper) {
547                 unsigned int ns_level = task_pid(father)->level;
548                 /*
549                  * Find the first ->is_child_subreaper ancestor in our pid_ns.
550                  * We can't check reaper != child_reaper to ensure we do not
551                  * cross the namespaces, the exiting parent could be injected
552                  * by setns() + fork().
553                  * We check pid->level, this is slightly more efficient than
554                  * task_active_pid_ns(reaper) != task_active_pid_ns(father).
555                  */
556                 for (reaper = father->real_parent;
557                      task_pid(reaper)->level == ns_level;
558                      reaper = reaper->real_parent) {
559                         if (reaper == &init_task)
560                                 break;
561                         if (!reaper->signal->is_child_subreaper)
562                                 continue;
563                         thread = find_alive_thread(reaper);
564                         if (thread)
565                                 return thread;
566                 }
567         }
568
569         return child_reaper;
570 }
571
572 /*
573 * Any that need to be release_task'd are put on the @dead list.
574  */
575 static void reparent_leader(struct task_struct *father, struct task_struct *p,
576                                 struct list_head *dead)
577 {
578         if (unlikely(p->exit_state == EXIT_DEAD))
579                 return;
580
581         /* We don't want people slaying init. */
582         p->exit_signal = SIGCHLD;
583
584         /* If it has exited notify the new parent about this child's death. */
585         if (!p->ptrace &&
586             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
587                 if (do_notify_parent(p, p->exit_signal)) {
588                         p->exit_state = EXIT_DEAD;
589                         list_add(&p->ptrace_entry, dead);
590                 }
591         }
592
593         kill_orphaned_pgrp(p, father);
594 }
595
596 /*
597  * This does two things:
598  *
599  * A.  Make init inherit all the child processes
600  * B.  Check to see if any process groups have become orphaned
601  *      as a result of our exiting, and if they have any stopped
602  *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
603  */
604 static void forget_original_parent(struct task_struct *father,
605                                         struct list_head *dead)
606 {
607         struct task_struct *p, *t, *reaper;
608
609         if (unlikely(!list_empty(&father->ptraced)))
610                 exit_ptrace(father, dead);
611
612         /* Can drop and reacquire tasklist_lock */
613         reaper = find_child_reaper(father, dead);
614         if (list_empty(&father->children))
615                 return;
616
617         reaper = find_new_reaper(father, reaper);
618         list_for_each_entry(p, &father->children, sibling) {
619                 for_each_thread(p, t) {
620                         RCU_INIT_POINTER(t->real_parent, reaper);
621                         BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
622                         if (likely(!t->ptrace))
623                                 t->parent = t->real_parent;
624                         if (t->pdeath_signal)
625                                 group_send_sig_info(t->pdeath_signal,
626                                                     SEND_SIG_NOINFO, t,
627                                                     PIDTYPE_TGID);
628                 }
629                 /*
630                  * If this is a threaded reparent there is no need to
631                  * notify anyone anything has happened.
632                  */
633                 if (!same_thread_group(reaper, father))
634                         reparent_leader(father, p, dead);
635         }
636         list_splice_tail_init(&father->children, &reaper->children);
637 }
638
639 /*
640  * Send signals to all our closest relatives so that they know
641  * to properly mourn us..
642  */
643 static void exit_notify(struct task_struct *tsk, int group_dead)
644 {
645         bool autoreap;
646         struct task_struct *p, *n;
647         LIST_HEAD(dead);
648
649         write_lock_irq(&tasklist_lock);
650         forget_original_parent(tsk, &dead);
651
652         if (group_dead)
653                 kill_orphaned_pgrp(tsk->group_leader, NULL);
654
655         tsk->exit_state = EXIT_ZOMBIE;
656         if (unlikely(tsk->ptrace)) {
657                 int sig = thread_group_leader(tsk) &&
658                                 thread_group_empty(tsk) &&
659                                 !ptrace_reparented(tsk) ?
660                         tsk->exit_signal : SIGCHLD;
661                 autoreap = do_notify_parent(tsk, sig);
662         } else if (thread_group_leader(tsk)) {
663                 autoreap = thread_group_empty(tsk) &&
664                         do_notify_parent(tsk, tsk->exit_signal);
665         } else {
666                 autoreap = true;
667         }
668
669         if (autoreap) {
670                 tsk->exit_state = EXIT_DEAD;
671                 list_add(&tsk->ptrace_entry, &dead);
672         }
673
674         /* mt-exec, de_thread() is waiting for group leader */
675         if (unlikely(tsk->signal->notify_count < 0))
676                 wake_up_process(tsk->signal->group_exit_task);
677         write_unlock_irq(&tasklist_lock);
678
679         list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
680                 list_del_init(&p->ptrace_entry);
681                 release_task(p);
682         }
683 }
684
685 #ifdef CONFIG_DEBUG_STACK_USAGE
686 static void check_stack_usage(void)
687 {
688         static DEFINE_SPINLOCK(low_water_lock);
689         static int lowest_to_date = THREAD_SIZE;
690         unsigned long free;
691
692         free = stack_not_used(current);
693
694         if (free >= lowest_to_date)
695                 return;
696
697         spin_lock(&low_water_lock);
698         if (free < lowest_to_date) {
699                 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
700                         current->comm, task_pid_nr(current), free);
701                 lowest_to_date = free;
702         }
703         spin_unlock(&low_water_lock);
704 }
705 #else
706 static inline void check_stack_usage(void) {}
707 #endif
708
709 void __noreturn do_exit(long code)
710 {
711         struct task_struct *tsk = current;
712         int group_dead;
713
714         /*
715          * We can get here from a kernel oops, sometimes with preemption off.
716          * Start by checking for critical errors.
717          * Then fix up important state like USER_DS and preemption.
718          * Then do everything else.
719          */
720
721         WARN_ON(blk_needs_flush_plug(tsk));
722
723         if (unlikely(in_interrupt()))
724                 panic("Aiee, killing interrupt handler!");
725         if (unlikely(!tsk->pid))
726                 panic("Attempted to kill the idle task!");
727
728         /*
729          * If do_exit is called because this processes oopsed, it's possible
730          * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
731          * continuing. Amongst other possible reasons, this is to prevent
732          * mm_release()->clear_child_tid() from writing to a user-controlled
733          * kernel address.
734          */
735         force_uaccess_begin();
736
737         if (unlikely(in_atomic())) {
738                 pr_info("note: %s[%d] exited with preempt_count %d\n",
739                         current->comm, task_pid_nr(current),
740                         preempt_count());
741                 preempt_count_set(PREEMPT_ENABLED);
742         }
743
744         profile_task_exit(tsk);
745         kcov_task_exit(tsk);
746
747         ptrace_event(PTRACE_EVENT_EXIT, code);
748
749         validate_creds_for_do_exit(tsk);
750
751         /*
752          * We're taking recursive faults here in do_exit. Safest is to just
753          * leave this task alone and wait for reboot.
754          */
755         if (unlikely(tsk->flags & PF_EXITING)) {
756                 pr_alert("Fixing recursive fault but reboot is needed!\n");
757                 futex_exit_recursive(tsk);
758                 set_current_state(TASK_UNINTERRUPTIBLE);
759                 schedule();
760         }
761
762         exit_signals(tsk);  /* sets PF_EXITING */
763
764         /* sync mm's RSS info before statistics gathering */
765         if (tsk->mm)
766                 sync_mm_rss(tsk->mm);
767         acct_update_integrals(tsk);
768         group_dead = atomic_dec_and_test(&tsk->signal->live);
769         if (group_dead) {
770                 /*
771                  * If the last thread of global init has exited, panic
772                  * immediately to get a useable coredump.
773                  */
774                 if (unlikely(is_global_init(tsk)))
775                         panic("Attempted to kill init! exitcode=0x%08x\n",
776                                 tsk->signal->group_exit_code ?: (int)code);
777
778 #ifdef CONFIG_POSIX_TIMERS
779                 hrtimer_cancel(&tsk->signal->real_timer);
780                 exit_itimers(tsk->signal);
781 #endif
782                 if (tsk->mm)
783                         setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
784         }
785         acct_collect(code, group_dead);
786         if (group_dead)
787                 tty_audit_exit();
788         audit_free(tsk);
789
790         tsk->exit_code = code;
791         taskstats_exit(tsk, group_dead);
792
793         exit_mm();
794
795         if (group_dead)
796                 acct_process();
797         trace_sched_process_exit(tsk);
798
799         exit_sem(tsk);
800         exit_shm(tsk);
801         exit_files(tsk);
802         exit_fs(tsk);
803         if (group_dead)
804                 disassociate_ctty(1);
805         exit_task_namespaces(tsk);
806         exit_task_work(tsk);
807         exit_thread(tsk);
808
809         /*
810          * Flush inherited counters to the parent - before the parent
811          * gets woken up by child-exit notifications.
812          *
813          * because of cgroup mode, must be called before cgroup_exit()
814          */
815         perf_event_exit_task(tsk);
816
817         sched_autogroup_exit_task(tsk);
818         cgroup_exit(tsk);
819
820         /*
821          * FIXME: do that only when needed, using sched_exit tracepoint
822          */
823         flush_ptrace_hw_breakpoint(tsk);
824
825         exit_tasks_rcu_start();
826         exit_notify(tsk, group_dead);
827         proc_exit_connector(tsk);
828         mpol_put_task_policy(tsk);
829 #ifdef CONFIG_FUTEX
830         if (unlikely(current->pi_state_cache))
831                 kfree(current->pi_state_cache);
832 #endif
833         /*
834          * Make sure we are holding no locks:
835          */
836         debug_check_no_locks_held();
837
838         if (tsk->io_context)
839                 exit_io_context(tsk);
840
841         if (tsk->splice_pipe)
842                 free_pipe_info(tsk->splice_pipe);
843
844         if (tsk->task_frag.page)
845                 put_page(tsk->task_frag.page);
846
847         validate_creds_for_do_exit(tsk);
848
849         check_stack_usage();
850         preempt_disable();
851         if (tsk->nr_dirtied)
852                 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
853         exit_rcu();
854         exit_tasks_rcu_finish();
855
856         lockdep_free_task(tsk);
857         do_task_dead();
858 }
859 EXPORT_SYMBOL_GPL(do_exit);
860
861 void complete_and_exit(struct completion *comp, long code)
862 {
863         if (comp)
864                 complete(comp);
865
866         do_exit(code);
867 }
868 EXPORT_SYMBOL(complete_and_exit);
869
870 SYSCALL_DEFINE1(exit, int, error_code)
871 {
872         do_exit((error_code&0xff)<<8);
873 }
874
875 /*
876  * Take down every thread in the group.  This is called by fatal signals
877  * as well as by sys_exit_group (below).
878  */
879 void
880 do_group_exit(int exit_code)
881 {
882         struct signal_struct *sig = current->signal;
883
884         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
885
886         if (signal_group_exit(sig))
887                 exit_code = sig->group_exit_code;
888         else if (!thread_group_empty(current)) {
889                 struct sighand_struct *const sighand = current->sighand;
890
891                 spin_lock_irq(&sighand->siglock);
892                 if (signal_group_exit(sig))
893                         /* Another thread got here before we took the lock.  */
894                         exit_code = sig->group_exit_code;
895                 else {
896                         sig->group_exit_code = exit_code;
897                         sig->flags = SIGNAL_GROUP_EXIT;
898                         zap_other_threads(current);
899                 }
900                 spin_unlock_irq(&sighand->siglock);
901         }
902
903         do_exit(exit_code);
904         /* NOTREACHED */
905 }
906
907 /*
908  * this kills every thread in the thread group. Note that any externally
909  * wait4()-ing process will get the correct exit code - even if this
910  * thread is not the thread group leader.
911  */
912 SYSCALL_DEFINE1(exit_group, int, error_code)
913 {
914         do_group_exit((error_code & 0xff) << 8);
915         /* NOTREACHED */
916         return 0;
917 }
918
919 struct waitid_info {
920         pid_t pid;
921         uid_t uid;
922         int status;
923         int cause;
924 };
925
926 struct wait_opts {
927         enum pid_type           wo_type;
928         int                     wo_flags;
929         struct pid              *wo_pid;
930
931         struct waitid_info      *wo_info;
932         int                     wo_stat;
933         struct rusage           *wo_rusage;
934
935         wait_queue_entry_t              child_wait;
936         int                     notask_error;
937 };
938
939 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
940 {
941         return  wo->wo_type == PIDTYPE_MAX ||
942                 task_pid_type(p, wo->wo_type) == wo->wo_pid;
943 }
944
945 static int
946 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
947 {
948         if (!eligible_pid(wo, p))
949                 return 0;
950
951         /*
952          * Wait for all children (clone and not) if __WALL is set or
953          * if it is traced by us.
954          */
955         if (ptrace || (wo->wo_flags & __WALL))
956                 return 1;
957
958         /*
959          * Otherwise, wait for clone children *only* if __WCLONE is set;
960          * otherwise, wait for non-clone children *only*.
961          *
962          * Note: a "clone" child here is one that reports to its parent
963          * using a signal other than SIGCHLD, or a non-leader thread which
964          * we can only see if it is traced by us.
965          */
966         if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
967                 return 0;
968
969         return 1;
970 }
971
972 /*
973  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
974  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
975  * the lock and this task is uninteresting.  If we return nonzero, we have
976  * released the lock and the system call should return.
977  */
978 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
979 {
980         int state, status;
981         pid_t pid = task_pid_vnr(p);
982         uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
983         struct waitid_info *infop;
984
985         if (!likely(wo->wo_flags & WEXITED))
986                 return 0;
987
988         if (unlikely(wo->wo_flags & WNOWAIT)) {
989                 status = p->exit_code;
990                 get_task_struct(p);
991                 read_unlock(&tasklist_lock);
992                 sched_annotate_sleep();
993                 if (wo->wo_rusage)
994                         getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
995                 put_task_struct(p);
996                 goto out_info;
997         }
998         /*
999          * Move the task's state to DEAD/TRACE, only one thread can do this.
1000          */
1001         state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1002                 EXIT_TRACE : EXIT_DEAD;
1003         if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1004                 return 0;
1005         /*
1006          * We own this thread, nobody else can reap it.
1007          */
1008         read_unlock(&tasklist_lock);
1009         sched_annotate_sleep();
1010
1011         /*
1012          * Check thread_group_leader() to exclude the traced sub-threads.
1013          */
1014         if (state == EXIT_DEAD && thread_group_leader(p)) {
1015                 struct signal_struct *sig = p->signal;
1016                 struct signal_struct *psig = current->signal;
1017                 unsigned long maxrss;
1018                 u64 tgutime, tgstime;
1019
1020                 /*
1021                  * The resource counters for the group leader are in its
1022                  * own task_struct.  Those for dead threads in the group
1023                  * are in its signal_struct, as are those for the child
1024                  * processes it has previously reaped.  All these
1025                  * accumulate in the parent's signal_struct c* fields.
1026                  *
1027                  * We don't bother to take a lock here to protect these
1028                  * p->signal fields because the whole thread group is dead
1029                  * and nobody can change them.
1030                  *
1031                  * psig->stats_lock also protects us from our sub-theads
1032                  * which can reap other children at the same time. Until
1033                  * we change k_getrusage()-like users to rely on this lock
1034                  * we have to take ->siglock as well.
1035                  *
1036                  * We use thread_group_cputime_adjusted() to get times for
1037                  * the thread group, which consolidates times for all threads
1038                  * in the group including the group leader.
1039                  */
1040                 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1041                 spin_lock_irq(&current->sighand->siglock);
1042                 write_seqlock(&psig->stats_lock);
1043                 psig->cutime += tgutime + sig->cutime;
1044                 psig->cstime += tgstime + sig->cstime;
1045                 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1046                 psig->cmin_flt +=
1047                         p->min_flt + sig->min_flt + sig->cmin_flt;
1048                 psig->cmaj_flt +=
1049                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1050                 psig->cnvcsw +=
1051                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1052                 psig->cnivcsw +=
1053                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1054                 psig->cinblock +=
1055                         task_io_get_inblock(p) +
1056                         sig->inblock + sig->cinblock;
1057                 psig->coublock +=
1058                         task_io_get_oublock(p) +
1059                         sig->oublock + sig->coublock;
1060                 maxrss = max(sig->maxrss, sig->cmaxrss);
1061                 if (psig->cmaxrss < maxrss)
1062                         psig->cmaxrss = maxrss;
1063                 task_io_accounting_add(&psig->ioac, &p->ioac);
1064                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1065                 write_sequnlock(&psig->stats_lock);
1066                 spin_unlock_irq(&current->sighand->siglock);
1067         }
1068
1069         if (wo->wo_rusage)
1070                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1071         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1072                 ? p->signal->group_exit_code : p->exit_code;
1073         wo->wo_stat = status;
1074
1075         if (state == EXIT_TRACE) {
1076                 write_lock_irq(&tasklist_lock);
1077                 /* We dropped tasklist, ptracer could die and untrace */
1078                 ptrace_unlink(p);
1079
1080                 /* If parent wants a zombie, don't release it now */
1081                 state = EXIT_ZOMBIE;
1082                 if (do_notify_parent(p, p->exit_signal))
1083                         state = EXIT_DEAD;
1084                 p->exit_state = state;
1085                 write_unlock_irq(&tasklist_lock);
1086         }
1087         if (state == EXIT_DEAD)
1088                 release_task(p);
1089
1090 out_info:
1091         infop = wo->wo_info;
1092         if (infop) {
1093                 if ((status & 0x7f) == 0) {
1094                         infop->cause = CLD_EXITED;
1095                         infop->status = status >> 8;
1096                 } else {
1097                         infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1098                         infop->status = status & 0x7f;
1099                 }
1100                 infop->pid = pid;
1101                 infop->uid = uid;
1102         }
1103
1104         return pid;
1105 }
1106
1107 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1108 {
1109         if (ptrace) {
1110                 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1111                         return &p->exit_code;
1112         } else {
1113                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1114                         return &p->signal->group_exit_code;
1115         }
1116         return NULL;
1117 }
1118
1119 /**
1120  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1121  * @wo: wait options
1122  * @ptrace: is the wait for ptrace
1123  * @p: task to wait for
1124  *
1125  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1126  *
1127  * CONTEXT:
1128  * read_lock(&tasklist_lock), which is released if return value is
1129  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1130  *
1131  * RETURNS:
1132  * 0 if wait condition didn't exist and search for other wait conditions
1133  * should continue.  Non-zero return, -errno on failure and @p's pid on
1134  * success, implies that tasklist_lock is released and wait condition
1135  * search should terminate.
1136  */
1137 static int wait_task_stopped(struct wait_opts *wo,
1138                                 int ptrace, struct task_struct *p)
1139 {
1140         struct waitid_info *infop;
1141         int exit_code, *p_code, why;
1142         uid_t uid = 0; /* unneeded, required by compiler */
1143         pid_t pid;
1144
1145         /*
1146          * Traditionally we see ptrace'd stopped tasks regardless of options.
1147          */
1148         if (!ptrace && !(wo->wo_flags & WUNTRACED))
1149                 return 0;
1150
1151         if (!task_stopped_code(p, ptrace))
1152                 return 0;
1153
1154         exit_code = 0;
1155         spin_lock_irq(&p->sighand->siglock);
1156
1157         p_code = task_stopped_code(p, ptrace);
1158         if (unlikely(!p_code))
1159                 goto unlock_sig;
1160
1161         exit_code = *p_code;
1162         if (!exit_code)
1163                 goto unlock_sig;
1164
1165         if (!unlikely(wo->wo_flags & WNOWAIT))
1166                 *p_code = 0;
1167
1168         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1169 unlock_sig:
1170         spin_unlock_irq(&p->sighand->siglock);
1171         if (!exit_code)
1172                 return 0;
1173
1174         /*
1175          * Now we are pretty sure this task is interesting.
1176          * Make sure it doesn't get reaped out from under us while we
1177          * give up the lock and then examine it below.  We don't want to
1178          * keep holding onto the tasklist_lock while we call getrusage and
1179          * possibly take page faults for user memory.
1180          */
1181         get_task_struct(p);
1182         pid = task_pid_vnr(p);
1183         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1184         read_unlock(&tasklist_lock);
1185         sched_annotate_sleep();
1186         if (wo->wo_rusage)
1187                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1188         put_task_struct(p);
1189
1190         if (likely(!(wo->wo_flags & WNOWAIT)))
1191                 wo->wo_stat = (exit_code << 8) | 0x7f;
1192
1193         infop = wo->wo_info;
1194         if (infop) {
1195                 infop->cause = why;
1196                 infop->status = exit_code;
1197                 infop->pid = pid;
1198                 infop->uid = uid;
1199         }
1200         return pid;
1201 }
1202
1203 /*
1204  * Handle do_wait work for one task in a live, non-stopped state.
1205  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1206  * the lock and this task is uninteresting.  If we return nonzero, we have
1207  * released the lock and the system call should return.
1208  */
1209 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1210 {
1211         struct waitid_info *infop;
1212         pid_t pid;
1213         uid_t uid;
1214
1215         if (!unlikely(wo->wo_flags & WCONTINUED))
1216                 return 0;
1217
1218         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1219                 return 0;
1220
1221         spin_lock_irq(&p->sighand->siglock);
1222         /* Re-check with the lock held.  */
1223         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1224                 spin_unlock_irq(&p->sighand->siglock);
1225                 return 0;
1226         }
1227         if (!unlikely(wo->wo_flags & WNOWAIT))
1228                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1229         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1230         spin_unlock_irq(&p->sighand->siglock);
1231
1232         pid = task_pid_vnr(p);
1233         get_task_struct(p);
1234         read_unlock(&tasklist_lock);
1235         sched_annotate_sleep();
1236         if (wo->wo_rusage)
1237                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1238         put_task_struct(p);
1239
1240         infop = wo->wo_info;
1241         if (!infop) {
1242                 wo->wo_stat = 0xffff;
1243         } else {
1244                 infop->cause = CLD_CONTINUED;
1245                 infop->pid = pid;
1246                 infop->uid = uid;
1247                 infop->status = SIGCONT;
1248         }
1249         return pid;
1250 }
1251
1252 /*
1253  * Consider @p for a wait by @parent.
1254  *
1255  * -ECHILD should be in ->notask_error before the first call.
1256  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1257  * Returns zero if the search for a child should continue;
1258  * then ->notask_error is 0 if @p is an eligible child,
1259  * or still -ECHILD.
1260  */
1261 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1262                                 struct task_struct *p)
1263 {
1264         /*
1265          * We can race with wait_task_zombie() from another thread.
1266          * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1267          * can't confuse the checks below.
1268          */
1269         int exit_state = READ_ONCE(p->exit_state);
1270         int ret;
1271
1272         if (unlikely(exit_state == EXIT_DEAD))
1273                 return 0;
1274
1275         ret = eligible_child(wo, ptrace, p);
1276         if (!ret)
1277                 return ret;
1278
1279         if (unlikely(exit_state == EXIT_TRACE)) {
1280                 /*
1281                  * ptrace == 0 means we are the natural parent. In this case
1282                  * we should clear notask_error, debugger will notify us.
1283                  */
1284                 if (likely(!ptrace))
1285                         wo->notask_error = 0;
1286                 return 0;
1287         }
1288
1289         if (likely(!ptrace) && unlikely(p->ptrace)) {
1290                 /*
1291                  * If it is traced by its real parent's group, just pretend
1292                  * the caller is ptrace_do_wait() and reap this child if it
1293                  * is zombie.
1294                  *
1295                  * This also hides group stop state from real parent; otherwise
1296                  * a single stop can be reported twice as group and ptrace stop.
1297                  * If a ptracer wants to distinguish these two events for its
1298                  * own children it should create a separate process which takes
1299                  * the role of real parent.
1300                  */
1301                 if (!ptrace_reparented(p))
1302                         ptrace = 1;
1303         }
1304
1305         /* slay zombie? */
1306         if (exit_state == EXIT_ZOMBIE) {
1307                 /* we don't reap group leaders with subthreads */
1308                 if (!delay_group_leader(p)) {
1309                         /*
1310                          * A zombie ptracee is only visible to its ptracer.
1311                          * Notification and reaping will be cascaded to the
1312                          * real parent when the ptracer detaches.
1313                          */
1314                         if (unlikely(ptrace) || likely(!p->ptrace))
1315                                 return wait_task_zombie(wo, p);
1316                 }
1317
1318                 /*
1319                  * Allow access to stopped/continued state via zombie by
1320                  * falling through.  Clearing of notask_error is complex.
1321                  *
1322                  * When !@ptrace:
1323                  *
1324                  * If WEXITED is set, notask_error should naturally be
1325                  * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1326                  * so, if there are live subthreads, there are events to
1327                  * wait for.  If all subthreads are dead, it's still safe
1328                  * to clear - this function will be called again in finite
1329                  * amount time once all the subthreads are released and
1330                  * will then return without clearing.
1331                  *
1332                  * When @ptrace:
1333                  *
1334                  * Stopped state is per-task and thus can't change once the
1335                  * target task dies.  Only continued and exited can happen.
1336                  * Clear notask_error if WCONTINUED | WEXITED.
1337                  */
1338                 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1339                         wo->notask_error = 0;
1340         } else {
1341                 /*
1342                  * @p is alive and it's gonna stop, continue or exit, so
1343                  * there always is something to wait for.
1344                  */
1345                 wo->notask_error = 0;
1346         }
1347
1348         /*
1349          * Wait for stopped.  Depending on @ptrace, different stopped state
1350          * is used and the two don't interact with each other.
1351          */
1352         ret = wait_task_stopped(wo, ptrace, p);
1353         if (ret)
1354                 return ret;
1355
1356         /*
1357          * Wait for continued.  There's only one continued state and the
1358          * ptracer can consume it which can confuse the real parent.  Don't
1359          * use WCONTINUED from ptracer.  You don't need or want it.
1360          */
1361         return wait_task_continued(wo, p);
1362 }
1363
1364 /*
1365  * Do the work of do_wait() for one thread in the group, @tsk.
1366  *
1367  * -ECHILD should be in ->notask_error before the first call.
1368  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1369  * Returns zero if the search for a child should continue; then
1370  * ->notask_error is 0 if there were any eligible children,
1371  * or still -ECHILD.
1372  */
1373 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1374 {
1375         struct task_struct *p;
1376
1377         list_for_each_entry(p, &tsk->children, sibling) {
1378                 int ret = wait_consider_task(wo, 0, p);
1379
1380                 if (ret)
1381                         return ret;
1382         }
1383
1384         return 0;
1385 }
1386
1387 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1388 {
1389         struct task_struct *p;
1390
1391         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1392                 int ret = wait_consider_task(wo, 1, p);
1393
1394                 if (ret)
1395                         return ret;
1396         }
1397
1398         return 0;
1399 }
1400
1401 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1402                                 int sync, void *key)
1403 {
1404         struct wait_opts *wo = container_of(wait, struct wait_opts,
1405                                                 child_wait);
1406         struct task_struct *p = key;
1407
1408         if (!eligible_pid(wo, p))
1409                 return 0;
1410
1411         if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1412                 return 0;
1413
1414         return default_wake_function(wait, mode, sync, key);
1415 }
1416
1417 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1418 {
1419         __wake_up_sync_key(&parent->signal->wait_chldexit,
1420                            TASK_INTERRUPTIBLE, p);
1421 }
1422
1423 static long do_wait(struct wait_opts *wo)
1424 {
1425         struct task_struct *tsk;
1426         int retval;
1427
1428         trace_sched_process_wait(wo->wo_pid);
1429
1430         init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1431         wo->child_wait.private = current;
1432         add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1433 repeat:
1434         /*
1435          * If there is nothing that can match our criteria, just get out.
1436          * We will clear ->notask_error to zero if we see any child that
1437          * might later match our criteria, even if we are not able to reap
1438          * it yet.
1439          */
1440         wo->notask_error = -ECHILD;
1441         if ((wo->wo_type < PIDTYPE_MAX) &&
1442            (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1443                 goto notask;
1444
1445         set_current_state(TASK_INTERRUPTIBLE);
1446         read_lock(&tasklist_lock);
1447         tsk = current;
1448         do {
1449                 retval = do_wait_thread(wo, tsk);
1450                 if (retval)
1451                         goto end;
1452
1453                 retval = ptrace_do_wait(wo, tsk);
1454                 if (retval)
1455                         goto end;
1456
1457                 if (wo->wo_flags & __WNOTHREAD)
1458                         break;
1459         } while_each_thread(current, tsk);
1460         read_unlock(&tasklist_lock);
1461
1462 notask:
1463         retval = wo->notask_error;
1464         if (!retval && !(wo->wo_flags & WNOHANG)) {
1465                 retval = -ERESTARTSYS;
1466                 if (!signal_pending(current)) {
1467                         schedule();
1468                         goto repeat;
1469                 }
1470         }
1471 end:
1472         __set_current_state(TASK_RUNNING);
1473         remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1474         return retval;
1475 }
1476
1477 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1478                           int options, struct rusage *ru)
1479 {
1480         struct wait_opts wo;
1481         struct pid *pid = NULL;
1482         enum pid_type type;
1483         long ret;
1484         unsigned int f_flags = 0;
1485
1486         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1487                         __WNOTHREAD|__WCLONE|__WALL))
1488                 return -EINVAL;
1489         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1490                 return -EINVAL;
1491
1492         switch (which) {
1493         case P_ALL:
1494                 type = PIDTYPE_MAX;
1495                 break;
1496         case P_PID:
1497                 type = PIDTYPE_PID;
1498                 if (upid <= 0)
1499                         return -EINVAL;
1500
1501                 pid = find_get_pid(upid);
1502                 break;
1503         case P_PGID:
1504                 type = PIDTYPE_PGID;
1505                 if (upid < 0)
1506                         return -EINVAL;
1507
1508                 if (upid)
1509                         pid = find_get_pid(upid);
1510                 else
1511                         pid = get_task_pid(current, PIDTYPE_PGID);
1512                 break;
1513         case P_PIDFD:
1514                 type = PIDTYPE_PID;
1515                 if (upid < 0)
1516                         return -EINVAL;
1517
1518                 pid = pidfd_get_pid(upid, &f_flags);
1519                 if (IS_ERR(pid))
1520                         return PTR_ERR(pid);
1521
1522                 break;
1523         default:
1524                 return -EINVAL;
1525         }
1526
1527         wo.wo_type      = type;
1528         wo.wo_pid       = pid;
1529         wo.wo_flags     = options;
1530         wo.wo_info      = infop;
1531         wo.wo_rusage    = ru;
1532         if (f_flags & O_NONBLOCK)
1533                 wo.wo_flags |= WNOHANG;
1534
1535         ret = do_wait(&wo);
1536         if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1537                 ret = -EAGAIN;
1538
1539         put_pid(pid);
1540         return ret;
1541 }
1542
1543 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1544                 infop, int, options, struct rusage __user *, ru)
1545 {
1546         struct rusage r;
1547         struct waitid_info info = {.status = 0};
1548         long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1549         int signo = 0;
1550
1551         if (err > 0) {
1552                 signo = SIGCHLD;
1553                 err = 0;
1554                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1555                         return -EFAULT;
1556         }
1557         if (!infop)
1558                 return err;
1559
1560         if (!user_write_access_begin(infop, sizeof(*infop)))
1561                 return -EFAULT;
1562
1563         unsafe_put_user(signo, &infop->si_signo, Efault);
1564         unsafe_put_user(0, &infop->si_errno, Efault);
1565         unsafe_put_user(info.cause, &infop->si_code, Efault);
1566         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1567         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1568         unsafe_put_user(info.status, &infop->si_status, Efault);
1569         user_write_access_end();
1570         return err;
1571 Efault:
1572         user_write_access_end();
1573         return -EFAULT;
1574 }
1575
1576 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1577                   struct rusage *ru)
1578 {
1579         struct wait_opts wo;
1580         struct pid *pid = NULL;
1581         enum pid_type type;
1582         long ret;
1583
1584         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1585                         __WNOTHREAD|__WCLONE|__WALL))
1586                 return -EINVAL;
1587
1588         /* -INT_MIN is not defined */
1589         if (upid == INT_MIN)
1590                 return -ESRCH;
1591
1592         if (upid == -1)
1593                 type = PIDTYPE_MAX;
1594         else if (upid < 0) {
1595                 type = PIDTYPE_PGID;
1596                 pid = find_get_pid(-upid);
1597         } else if (upid == 0) {
1598                 type = PIDTYPE_PGID;
1599                 pid = get_task_pid(current, PIDTYPE_PGID);
1600         } else /* upid > 0 */ {
1601                 type = PIDTYPE_PID;
1602                 pid = find_get_pid(upid);
1603         }
1604
1605         wo.wo_type      = type;
1606         wo.wo_pid       = pid;
1607         wo.wo_flags     = options | WEXITED;
1608         wo.wo_info      = NULL;
1609         wo.wo_stat      = 0;
1610         wo.wo_rusage    = ru;
1611         ret = do_wait(&wo);
1612         put_pid(pid);
1613         if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1614                 ret = -EFAULT;
1615
1616         return ret;
1617 }
1618
1619 int kernel_wait(pid_t pid, int *stat)
1620 {
1621         struct wait_opts wo = {
1622                 .wo_type        = PIDTYPE_PID,
1623                 .wo_pid         = find_get_pid(pid),
1624                 .wo_flags       = WEXITED,
1625         };
1626         int ret;
1627
1628         ret = do_wait(&wo);
1629         if (ret > 0 && wo.wo_stat)
1630                 *stat = wo.wo_stat;
1631         put_pid(wo.wo_pid);
1632         return ret;
1633 }
1634
1635 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1636                 int, options, struct rusage __user *, ru)
1637 {
1638         struct rusage r;
1639         long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1640
1641         if (err > 0) {
1642                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1643                         return -EFAULT;
1644         }
1645         return err;
1646 }
1647
1648 #ifdef __ARCH_WANT_SYS_WAITPID
1649
1650 /*
1651  * sys_waitpid() remains for compatibility. waitpid() should be
1652  * implemented by calling sys_wait4() from libc.a.
1653  */
1654 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1655 {
1656         return kernel_wait4(pid, stat_addr, options, NULL);
1657 }
1658
1659 #endif
1660
1661 #ifdef CONFIG_COMPAT
1662 COMPAT_SYSCALL_DEFINE4(wait4,
1663         compat_pid_t, pid,
1664         compat_uint_t __user *, stat_addr,
1665         int, options,
1666         struct compat_rusage __user *, ru)
1667 {
1668         struct rusage r;
1669         long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1670         if (err > 0) {
1671                 if (ru && put_compat_rusage(&r, ru))
1672                         return -EFAULT;
1673         }
1674         return err;
1675 }
1676
1677 COMPAT_SYSCALL_DEFINE5(waitid,
1678                 int, which, compat_pid_t, pid,
1679                 struct compat_siginfo __user *, infop, int, options,
1680                 struct compat_rusage __user *, uru)
1681 {
1682         struct rusage ru;
1683         struct waitid_info info = {.status = 0};
1684         long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1685         int signo = 0;
1686         if (err > 0) {
1687                 signo = SIGCHLD;
1688                 err = 0;
1689                 if (uru) {
1690                         /* kernel_waitid() overwrites everything in ru */
1691                         if (COMPAT_USE_64BIT_TIME)
1692                                 err = copy_to_user(uru, &ru, sizeof(ru));
1693                         else
1694                                 err = put_compat_rusage(&ru, uru);
1695                         if (err)
1696                                 return -EFAULT;
1697                 }
1698         }
1699
1700         if (!infop)
1701                 return err;
1702
1703         if (!user_write_access_begin(infop, sizeof(*infop)))
1704                 return -EFAULT;
1705
1706         unsafe_put_user(signo, &infop->si_signo, Efault);
1707         unsafe_put_user(0, &infop->si_errno, Efault);
1708         unsafe_put_user(info.cause, &infop->si_code, Efault);
1709         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1710         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1711         unsafe_put_user(info.status, &infop->si_status, Efault);
1712         user_write_access_end();
1713         return err;
1714 Efault:
1715         user_write_access_end();
1716         return -EFAULT;
1717 }
1718 #endif
1719
1720 /**
1721  * thread_group_exited - check that a thread group has exited
1722  * @pid: tgid of thread group to be checked.
1723  *
1724  * Test if the thread group represented by tgid has exited (all
1725  * threads are zombies, dead or completely gone).
1726  *
1727  * Return: true if the thread group has exited. false otherwise.
1728  */
1729 bool thread_group_exited(struct pid *pid)
1730 {
1731         struct task_struct *task;
1732         bool exited;
1733
1734         rcu_read_lock();
1735         task = pid_task(pid, PIDTYPE_PID);
1736         exited = !task ||
1737                 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1738         rcu_read_unlock();
1739
1740         return exited;
1741 }
1742 EXPORT_SYMBOL(thread_group_exited);
1743
1744 __weak void abort(void)
1745 {
1746         BUG();
1747
1748         /* if that doesn't kill us, halt */
1749         panic("Oops failed to kill thread");
1750 }
1751 EXPORT_SYMBOL(abort);