Merge tag 'for-linus-5.15b-rc3-tag' of git://git.kernel.org/pub/scm/linux/kernel...
[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 #include <linux/io_uring.h>
67
68 #include <linux/uaccess.h>
69 #include <asm/unistd.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 *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 #endif
109
110         if (group_dead) {
111                 tty = sig->tty;
112                 sig->tty = NULL;
113         } else {
114                 /*
115                  * If there is any task waiting for the group exit
116                  * then notify it:
117                  */
118                 if (sig->notify_count > 0 && !--sig->notify_count)
119                         wake_up_process(sig->group_exit_task);
120
121                 if (tsk == sig->curr_target)
122                         sig->curr_target = next_thread(tsk);
123         }
124
125         add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
126                               sizeof(unsigned long long));
127
128         /*
129          * Accumulate here the counters for all threads as they die. We could
130          * skip the group leader because it is the last user of signal_struct,
131          * but we want to avoid the race with thread_group_cputime() which can
132          * see the empty ->thread_head list.
133          */
134         task_cputime(tsk, &utime, &stime);
135         write_seqlock(&sig->stats_lock);
136         sig->utime += utime;
137         sig->stime += stime;
138         sig->gtime += task_gtime(tsk);
139         sig->min_flt += tsk->min_flt;
140         sig->maj_flt += tsk->maj_flt;
141         sig->nvcsw += tsk->nvcsw;
142         sig->nivcsw += tsk->nivcsw;
143         sig->inblock += task_io_get_inblock(tsk);
144         sig->oublock += task_io_get_oublock(tsk);
145         task_io_accounting_add(&sig->ioac, &tsk->ioac);
146         sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
147         sig->nr_threads--;
148         __unhash_process(tsk, group_dead);
149         write_sequnlock(&sig->stats_lock);
150
151         /*
152          * Do this under ->siglock, we can race with another thread
153          * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
154          */
155         flush_sigqueue(&tsk->pending);
156         tsk->sighand = NULL;
157         spin_unlock(&sighand->siglock);
158
159         __cleanup_sighand(sighand);
160         clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
161         if (group_dead) {
162                 flush_sigqueue(&sig->shared_pending);
163                 tty_kref_put(tty);
164         }
165 }
166
167 static void delayed_put_task_struct(struct rcu_head *rhp)
168 {
169         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
170
171         perf_event_delayed_put(tsk);
172         trace_sched_process_free(tsk);
173         put_task_struct(tsk);
174 }
175
176 void put_task_struct_rcu_user(struct task_struct *task)
177 {
178         if (refcount_dec_and_test(&task->rcu_users))
179                 call_rcu(&task->rcu, delayed_put_task_struct);
180 }
181
182 void release_task(struct task_struct *p)
183 {
184         struct task_struct *leader;
185         struct pid *thread_pid;
186         int zap_leader;
187 repeat:
188         /* don't need to get the RCU readlock here - the process is dead and
189          * can't be modifying its own credentials. But shut RCU-lockdep up */
190         rcu_read_lock();
191         dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
192         rcu_read_unlock();
193
194         cgroup_release(p);
195
196         write_lock_irq(&tasklist_lock);
197         ptrace_release_task(p);
198         thread_pid = get_pid(p->thread_pid);
199         __exit_signal(p);
200
201         /*
202          * If we are the last non-leader member of the thread
203          * group, and the leader is zombie, then notify the
204          * group leader's parent process. (if it wants notification.)
205          */
206         zap_leader = 0;
207         leader = p->group_leader;
208         if (leader != p && thread_group_empty(leader)
209                         && leader->exit_state == EXIT_ZOMBIE) {
210                 /*
211                  * If we were the last child thread and the leader has
212                  * exited already, and the leader's parent ignores SIGCHLD,
213                  * then we are the one who should release the leader.
214                  */
215                 zap_leader = do_notify_parent(leader, leader->exit_signal);
216                 if (zap_leader)
217                         leader->exit_state = EXIT_DEAD;
218         }
219
220         write_unlock_irq(&tasklist_lock);
221         seccomp_filter_release(p);
222         proc_flush_pid(thread_pid);
223         put_pid(thread_pid);
224         release_thread(p);
225         put_task_struct_rcu_user(p);
226
227         p = leader;
228         if (unlikely(zap_leader))
229                 goto repeat;
230 }
231
232 int rcuwait_wake_up(struct rcuwait *w)
233 {
234         int ret = 0;
235         struct task_struct *task;
236
237         rcu_read_lock();
238
239         /*
240          * Order condition vs @task, such that everything prior to the load
241          * of @task is visible. This is the condition as to why the user called
242          * rcuwait_wake() in the first place. Pairs with set_current_state()
243          * barrier (A) in rcuwait_wait_event().
244          *
245          *    WAIT                WAKE
246          *    [S] tsk = current   [S] cond = true
247          *        MB (A)              MB (B)
248          *    [L] cond            [L] tsk
249          */
250         smp_mb(); /* (B) */
251
252         task = rcu_dereference(w->task);
253         if (task)
254                 ret = wake_up_process(task);
255         rcu_read_unlock();
256
257         return ret;
258 }
259 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
260
261 /*
262  * Determine if a process group is "orphaned", according to the POSIX
263  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
264  * by terminal-generated stop signals.  Newly orphaned process groups are
265  * to receive a SIGHUP and a SIGCONT.
266  *
267  * "I ask you, have you ever known what it is to be an orphan?"
268  */
269 static int will_become_orphaned_pgrp(struct pid *pgrp,
270                                         struct task_struct *ignored_task)
271 {
272         struct task_struct *p;
273
274         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
275                 if ((p == ignored_task) ||
276                     (p->exit_state && thread_group_empty(p)) ||
277                     is_global_init(p->real_parent))
278                         continue;
279
280                 if (task_pgrp(p->real_parent) != pgrp &&
281                     task_session(p->real_parent) == task_session(p))
282                         return 0;
283         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
284
285         return 1;
286 }
287
288 int is_current_pgrp_orphaned(void)
289 {
290         int retval;
291
292         read_lock(&tasklist_lock);
293         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
294         read_unlock(&tasklist_lock);
295
296         return retval;
297 }
298
299 static bool has_stopped_jobs(struct pid *pgrp)
300 {
301         struct task_struct *p;
302
303         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
304                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
305                         return true;
306         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
307
308         return false;
309 }
310
311 /*
312  * Check to see if any process groups have become orphaned as
313  * a result of our exiting, and if they have any stopped jobs,
314  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
315  */
316 static void
317 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
318 {
319         struct pid *pgrp = task_pgrp(tsk);
320         struct task_struct *ignored_task = tsk;
321
322         if (!parent)
323                 /* exit: our father is in a different pgrp than
324                  * we are and we were the only connection outside.
325                  */
326                 parent = tsk->real_parent;
327         else
328                 /* reparent: our child is in a different pgrp than
329                  * we are, and it was the only connection outside.
330                  */
331                 ignored_task = NULL;
332
333         if (task_pgrp(parent) != pgrp &&
334             task_session(parent) == task_session(tsk) &&
335             will_become_orphaned_pgrp(pgrp, ignored_task) &&
336             has_stopped_jobs(pgrp)) {
337                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
338                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
339         }
340 }
341
342 #ifdef CONFIG_MEMCG
343 /*
344  * A task is exiting.   If it owned this mm, find a new owner for the mm.
345  */
346 void mm_update_next_owner(struct mm_struct *mm)
347 {
348         struct task_struct *c, *g, *p = current;
349
350 retry:
351         /*
352          * If the exiting or execing task is not the owner, it's
353          * someone else's problem.
354          */
355         if (mm->owner != p)
356                 return;
357         /*
358          * The current owner is exiting/execing and there are no other
359          * candidates.  Do not leave the mm pointing to a possibly
360          * freed task structure.
361          */
362         if (atomic_read(&mm->mm_users) <= 1) {
363                 WRITE_ONCE(mm->owner, NULL);
364                 return;
365         }
366
367         read_lock(&tasklist_lock);
368         /*
369          * Search in the children
370          */
371         list_for_each_entry(c, &p->children, sibling) {
372                 if (c->mm == mm)
373                         goto assign_new_owner;
374         }
375
376         /*
377          * Search in the siblings
378          */
379         list_for_each_entry(c, &p->real_parent->children, sibling) {
380                 if (c->mm == mm)
381                         goto assign_new_owner;
382         }
383
384         /*
385          * Search through everything else, we should not get here often.
386          */
387         for_each_process(g) {
388                 if (g->flags & PF_KTHREAD)
389                         continue;
390                 for_each_thread(g, c) {
391                         if (c->mm == mm)
392                                 goto assign_new_owner;
393                         if (c->mm)
394                                 break;
395                 }
396         }
397         read_unlock(&tasklist_lock);
398         /*
399          * We found no owner yet mm_users > 1: this implies that we are
400          * most likely racing with swapoff (try_to_unuse()) or /proc or
401          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
402          */
403         WRITE_ONCE(mm->owner, NULL);
404         return;
405
406 assign_new_owner:
407         BUG_ON(c == p);
408         get_task_struct(c);
409         /*
410          * The task_lock protects c->mm from changing.
411          * We always want mm->owner->mm == mm
412          */
413         task_lock(c);
414         /*
415          * Delay read_unlock() till we have the task_lock()
416          * to ensure that c does not slip away underneath us
417          */
418         read_unlock(&tasklist_lock);
419         if (c->mm != mm) {
420                 task_unlock(c);
421                 put_task_struct(c);
422                 goto retry;
423         }
424         WRITE_ONCE(mm->owner, c);
425         task_unlock(c);
426         put_task_struct(c);
427 }
428 #endif /* CONFIG_MEMCG */
429
430 /*
431  * Turn us into a lazy TLB process if we
432  * aren't already..
433  */
434 static void exit_mm(void)
435 {
436         struct mm_struct *mm = current->mm;
437         struct core_state *core_state;
438
439         exit_mm_release(current, mm);
440         if (!mm)
441                 return;
442         sync_mm_rss(mm);
443         /*
444          * Serialize with any possible pending coredump.
445          * We must hold mmap_lock around checking core_state
446          * and clearing tsk->mm.  The core-inducing thread
447          * will increment ->nr_threads for each thread in the
448          * group with ->mm != NULL.
449          */
450         mmap_read_lock(mm);
451         core_state = mm->core_state;
452         if (core_state) {
453                 struct core_thread self;
454
455                 mmap_read_unlock(mm);
456
457                 self.task = current;
458                 if (self.task->flags & PF_SIGNALED)
459                         self.next = xchg(&core_state->dumper.next, &self);
460                 else
461                         self.task = NULL;
462                 /*
463                  * Implies mb(), the result of xchg() must be visible
464                  * to core_state->dumper.
465                  */
466                 if (atomic_dec_and_test(&core_state->nr_threads))
467                         complete(&core_state->startup);
468
469                 for (;;) {
470                         set_current_state(TASK_UNINTERRUPTIBLE);
471                         if (!self.task) /* see coredump_finish() */
472                                 break;
473                         freezable_schedule();
474                 }
475                 __set_current_state(TASK_RUNNING);
476                 mmap_read_lock(mm);
477         }
478         mmgrab(mm);
479         BUG_ON(mm != current->active_mm);
480         /* more a memory barrier than a real lock */
481         task_lock(current);
482         /*
483          * When a thread stops operating on an address space, the loop
484          * in membarrier_private_expedited() may not observe that
485          * tsk->mm, and the loop in membarrier_global_expedited() may
486          * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
487          * rq->membarrier_state, so those would not issue an IPI.
488          * Membarrier requires a memory barrier after accessing
489          * user-space memory, before clearing tsk->mm or the
490          * rq->membarrier_state.
491          */
492         smp_mb__after_spinlock();
493         local_irq_disable();
494         current->mm = NULL;
495         membarrier_update_current_mm(NULL);
496         enter_lazy_tlb(mm, current);
497         local_irq_enable();
498         task_unlock(current);
499         mmap_read_unlock(mm);
500         mm_update_next_owner(mm);
501         mmput(mm);
502         if (test_thread_flag(TIF_MEMDIE))
503                 exit_oom_victim();
504 }
505
506 static struct task_struct *find_alive_thread(struct task_struct *p)
507 {
508         struct task_struct *t;
509
510         for_each_thread(p, t) {
511                 if (!(t->flags & PF_EXITING))
512                         return t;
513         }
514         return NULL;
515 }
516
517 static struct task_struct *find_child_reaper(struct task_struct *father,
518                                                 struct list_head *dead)
519         __releases(&tasklist_lock)
520         __acquires(&tasklist_lock)
521 {
522         struct pid_namespace *pid_ns = task_active_pid_ns(father);
523         struct task_struct *reaper = pid_ns->child_reaper;
524         struct task_struct *p, *n;
525
526         if (likely(reaper != father))
527                 return reaper;
528
529         reaper = find_alive_thread(father);
530         if (reaper) {
531                 pid_ns->child_reaper = reaper;
532                 return reaper;
533         }
534
535         write_unlock_irq(&tasklist_lock);
536
537         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
538                 list_del_init(&p->ptrace_entry);
539                 release_task(p);
540         }
541
542         zap_pid_ns_processes(pid_ns);
543         write_lock_irq(&tasklist_lock);
544
545         return father;
546 }
547
548 /*
549  * When we die, we re-parent all our children, and try to:
550  * 1. give them to another thread in our thread group, if such a member exists
551  * 2. give it to the first ancestor process which prctl'd itself as a
552  *    child_subreaper for its children (like a service manager)
553  * 3. give it to the init process (PID 1) in our pid namespace
554  */
555 static struct task_struct *find_new_reaper(struct task_struct *father,
556                                            struct task_struct *child_reaper)
557 {
558         struct task_struct *thread, *reaper;
559
560         thread = find_alive_thread(father);
561         if (thread)
562                 return thread;
563
564         if (father->signal->has_child_subreaper) {
565                 unsigned int ns_level = task_pid(father)->level;
566                 /*
567                  * Find the first ->is_child_subreaper ancestor in our pid_ns.
568                  * We can't check reaper != child_reaper to ensure we do not
569                  * cross the namespaces, the exiting parent could be injected
570                  * by setns() + fork().
571                  * We check pid->level, this is slightly more efficient than
572                  * task_active_pid_ns(reaper) != task_active_pid_ns(father).
573                  */
574                 for (reaper = father->real_parent;
575                      task_pid(reaper)->level == ns_level;
576                      reaper = reaper->real_parent) {
577                         if (reaper == &init_task)
578                                 break;
579                         if (!reaper->signal->is_child_subreaper)
580                                 continue;
581                         thread = find_alive_thread(reaper);
582                         if (thread)
583                                 return thread;
584                 }
585         }
586
587         return child_reaper;
588 }
589
590 /*
591 * Any that need to be release_task'd are put on the @dead list.
592  */
593 static void reparent_leader(struct task_struct *father, struct task_struct *p,
594                                 struct list_head *dead)
595 {
596         if (unlikely(p->exit_state == EXIT_DEAD))
597                 return;
598
599         /* We don't want people slaying init. */
600         p->exit_signal = SIGCHLD;
601
602         /* If it has exited notify the new parent about this child's death. */
603         if (!p->ptrace &&
604             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
605                 if (do_notify_parent(p, p->exit_signal)) {
606                         p->exit_state = EXIT_DEAD;
607                         list_add(&p->ptrace_entry, dead);
608                 }
609         }
610
611         kill_orphaned_pgrp(p, father);
612 }
613
614 /*
615  * This does two things:
616  *
617  * A.  Make init inherit all the child processes
618  * B.  Check to see if any process groups have become orphaned
619  *      as a result of our exiting, and if they have any stopped
620  *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
621  */
622 static void forget_original_parent(struct task_struct *father,
623                                         struct list_head *dead)
624 {
625         struct task_struct *p, *t, *reaper;
626
627         if (unlikely(!list_empty(&father->ptraced)))
628                 exit_ptrace(father, dead);
629
630         /* Can drop and reacquire tasklist_lock */
631         reaper = find_child_reaper(father, dead);
632         if (list_empty(&father->children))
633                 return;
634
635         reaper = find_new_reaper(father, reaper);
636         list_for_each_entry(p, &father->children, sibling) {
637                 for_each_thread(p, t) {
638                         RCU_INIT_POINTER(t->real_parent, reaper);
639                         BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
640                         if (likely(!t->ptrace))
641                                 t->parent = t->real_parent;
642                         if (t->pdeath_signal)
643                                 group_send_sig_info(t->pdeath_signal,
644                                                     SEND_SIG_NOINFO, t,
645                                                     PIDTYPE_TGID);
646                 }
647                 /*
648                  * If this is a threaded reparent there is no need to
649                  * notify anyone anything has happened.
650                  */
651                 if (!same_thread_group(reaper, father))
652                         reparent_leader(father, p, dead);
653         }
654         list_splice_tail_init(&father->children, &reaper->children);
655 }
656
657 /*
658  * Send signals to all our closest relatives so that they know
659  * to properly mourn us..
660  */
661 static void exit_notify(struct task_struct *tsk, int group_dead)
662 {
663         bool autoreap;
664         struct task_struct *p, *n;
665         LIST_HEAD(dead);
666
667         write_lock_irq(&tasklist_lock);
668         forget_original_parent(tsk, &dead);
669
670         if (group_dead)
671                 kill_orphaned_pgrp(tsk->group_leader, NULL);
672
673         tsk->exit_state = EXIT_ZOMBIE;
674         if (unlikely(tsk->ptrace)) {
675                 int sig = thread_group_leader(tsk) &&
676                                 thread_group_empty(tsk) &&
677                                 !ptrace_reparented(tsk) ?
678                         tsk->exit_signal : SIGCHLD;
679                 autoreap = do_notify_parent(tsk, sig);
680         } else if (thread_group_leader(tsk)) {
681                 autoreap = thread_group_empty(tsk) &&
682                         do_notify_parent(tsk, tsk->exit_signal);
683         } else {
684                 autoreap = true;
685         }
686
687         if (autoreap) {
688                 tsk->exit_state = EXIT_DEAD;
689                 list_add(&tsk->ptrace_entry, &dead);
690         }
691
692         /* mt-exec, de_thread() is waiting for group leader */
693         if (unlikely(tsk->signal->notify_count < 0))
694                 wake_up_process(tsk->signal->group_exit_task);
695         write_unlock_irq(&tasklist_lock);
696
697         list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
698                 list_del_init(&p->ptrace_entry);
699                 release_task(p);
700         }
701 }
702
703 #ifdef CONFIG_DEBUG_STACK_USAGE
704 static void check_stack_usage(void)
705 {
706         static DEFINE_SPINLOCK(low_water_lock);
707         static int lowest_to_date = THREAD_SIZE;
708         unsigned long free;
709
710         free = stack_not_used(current);
711
712         if (free >= lowest_to_date)
713                 return;
714
715         spin_lock(&low_water_lock);
716         if (free < lowest_to_date) {
717                 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
718                         current->comm, task_pid_nr(current), free);
719                 lowest_to_date = free;
720         }
721         spin_unlock(&low_water_lock);
722 }
723 #else
724 static inline void check_stack_usage(void) {}
725 #endif
726
727 void __noreturn do_exit(long code)
728 {
729         struct task_struct *tsk = current;
730         int group_dead;
731
732         /*
733          * We can get here from a kernel oops, sometimes with preemption off.
734          * Start by checking for critical errors.
735          * Then fix up important state like USER_DS and preemption.
736          * Then do everything else.
737          */
738
739         WARN_ON(blk_needs_flush_plug(tsk));
740
741         if (unlikely(in_interrupt()))
742                 panic("Aiee, killing interrupt handler!");
743         if (unlikely(!tsk->pid))
744                 panic("Attempted to kill the idle task!");
745
746         /*
747          * If do_exit is called because this processes oopsed, it's possible
748          * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
749          * continuing. Amongst other possible reasons, this is to prevent
750          * mm_release()->clear_child_tid() from writing to a user-controlled
751          * kernel address.
752          */
753         force_uaccess_begin();
754
755         if (unlikely(in_atomic())) {
756                 pr_info("note: %s[%d] exited with preempt_count %d\n",
757                         current->comm, task_pid_nr(current),
758                         preempt_count());
759                 preempt_count_set(PREEMPT_ENABLED);
760         }
761
762         profile_task_exit(tsk);
763         kcov_task_exit(tsk);
764
765         ptrace_event(PTRACE_EVENT_EXIT, code);
766
767         validate_creds_for_do_exit(tsk);
768
769         /*
770          * We're taking recursive faults here in do_exit. Safest is to just
771          * leave this task alone and wait for reboot.
772          */
773         if (unlikely(tsk->flags & PF_EXITING)) {
774                 pr_alert("Fixing recursive fault but reboot is needed!\n");
775                 futex_exit_recursive(tsk);
776                 set_current_state(TASK_UNINTERRUPTIBLE);
777                 schedule();
778         }
779
780         io_uring_files_cancel();
781         exit_signals(tsk);  /* sets PF_EXITING */
782
783         /* sync mm's RSS info before statistics gathering */
784         if (tsk->mm)
785                 sync_mm_rss(tsk->mm);
786         acct_update_integrals(tsk);
787         group_dead = atomic_dec_and_test(&tsk->signal->live);
788         if (group_dead) {
789                 /*
790                  * If the last thread of global init has exited, panic
791                  * immediately to get a useable coredump.
792                  */
793                 if (unlikely(is_global_init(tsk)))
794                         panic("Attempted to kill init! exitcode=0x%08x\n",
795                                 tsk->signal->group_exit_code ?: (int)code);
796
797 #ifdef CONFIG_POSIX_TIMERS
798                 hrtimer_cancel(&tsk->signal->real_timer);
799                 exit_itimers(tsk->signal);
800 #endif
801                 if (tsk->mm)
802                         setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
803         }
804         acct_collect(code, group_dead);
805         if (group_dead)
806                 tty_audit_exit();
807         audit_free(tsk);
808
809         tsk->exit_code = code;
810         taskstats_exit(tsk, group_dead);
811
812         exit_mm();
813
814         if (group_dead)
815                 acct_process();
816         trace_sched_process_exit(tsk);
817
818         exit_sem(tsk);
819         exit_shm(tsk);
820         exit_files(tsk);
821         exit_fs(tsk);
822         if (group_dead)
823                 disassociate_ctty(1);
824         exit_task_namespaces(tsk);
825         exit_task_work(tsk);
826         exit_thread(tsk);
827
828         /*
829          * Flush inherited counters to the parent - before the parent
830          * gets woken up by child-exit notifications.
831          *
832          * because of cgroup mode, must be called before cgroup_exit()
833          */
834         perf_event_exit_task(tsk);
835
836         sched_autogroup_exit_task(tsk);
837         cgroup_exit(tsk);
838
839         /*
840          * FIXME: do that only when needed, using sched_exit tracepoint
841          */
842         flush_ptrace_hw_breakpoint(tsk);
843
844         exit_tasks_rcu_start();
845         exit_notify(tsk, group_dead);
846         proc_exit_connector(tsk);
847         mpol_put_task_policy(tsk);
848 #ifdef CONFIG_FUTEX
849         if (unlikely(current->pi_state_cache))
850                 kfree(current->pi_state_cache);
851 #endif
852         /*
853          * Make sure we are holding no locks:
854          */
855         debug_check_no_locks_held();
856
857         if (tsk->io_context)
858                 exit_io_context(tsk);
859
860         if (tsk->splice_pipe)
861                 free_pipe_info(tsk->splice_pipe);
862
863         if (tsk->task_frag.page)
864                 put_page(tsk->task_frag.page);
865
866         validate_creds_for_do_exit(tsk);
867
868         check_stack_usage();
869         preempt_disable();
870         if (tsk->nr_dirtied)
871                 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
872         exit_rcu();
873         exit_tasks_rcu_finish();
874
875         lockdep_free_task(tsk);
876         do_task_dead();
877 }
878 EXPORT_SYMBOL_GPL(do_exit);
879
880 void complete_and_exit(struct completion *comp, long code)
881 {
882         if (comp)
883                 complete(comp);
884
885         do_exit(code);
886 }
887 EXPORT_SYMBOL(complete_and_exit);
888
889 SYSCALL_DEFINE1(exit, int, error_code)
890 {
891         do_exit((error_code&0xff)<<8);
892 }
893
894 /*
895  * Take down every thread in the group.  This is called by fatal signals
896  * as well as by sys_exit_group (below).
897  */
898 void
899 do_group_exit(int exit_code)
900 {
901         struct signal_struct *sig = current->signal;
902
903         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
904
905         if (signal_group_exit(sig))
906                 exit_code = sig->group_exit_code;
907         else if (!thread_group_empty(current)) {
908                 struct sighand_struct *const sighand = current->sighand;
909
910                 spin_lock_irq(&sighand->siglock);
911                 if (signal_group_exit(sig))
912                         /* Another thread got here before we took the lock.  */
913                         exit_code = sig->group_exit_code;
914                 else {
915                         sig->group_exit_code = exit_code;
916                         sig->flags = SIGNAL_GROUP_EXIT;
917                         zap_other_threads(current);
918                 }
919                 spin_unlock_irq(&sighand->siglock);
920         }
921
922         do_exit(exit_code);
923         /* NOTREACHED */
924 }
925
926 /*
927  * this kills every thread in the thread group. Note that any externally
928  * wait4()-ing process will get the correct exit code - even if this
929  * thread is not the thread group leader.
930  */
931 SYSCALL_DEFINE1(exit_group, int, error_code)
932 {
933         do_group_exit((error_code & 0xff) << 8);
934         /* NOTREACHED */
935         return 0;
936 }
937
938 struct waitid_info {
939         pid_t pid;
940         uid_t uid;
941         int status;
942         int cause;
943 };
944
945 struct wait_opts {
946         enum pid_type           wo_type;
947         int                     wo_flags;
948         struct pid              *wo_pid;
949
950         struct waitid_info      *wo_info;
951         int                     wo_stat;
952         struct rusage           *wo_rusage;
953
954         wait_queue_entry_t              child_wait;
955         int                     notask_error;
956 };
957
958 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
959 {
960         return  wo->wo_type == PIDTYPE_MAX ||
961                 task_pid_type(p, wo->wo_type) == wo->wo_pid;
962 }
963
964 static int
965 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
966 {
967         if (!eligible_pid(wo, p))
968                 return 0;
969
970         /*
971          * Wait for all children (clone and not) if __WALL is set or
972          * if it is traced by us.
973          */
974         if (ptrace || (wo->wo_flags & __WALL))
975                 return 1;
976
977         /*
978          * Otherwise, wait for clone children *only* if __WCLONE is set;
979          * otherwise, wait for non-clone children *only*.
980          *
981          * Note: a "clone" child here is one that reports to its parent
982          * using a signal other than SIGCHLD, or a non-leader thread which
983          * we can only see if it is traced by us.
984          */
985         if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
986                 return 0;
987
988         return 1;
989 }
990
991 /*
992  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
993  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
994  * the lock and this task is uninteresting.  If we return nonzero, we have
995  * released the lock and the system call should return.
996  */
997 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
998 {
999         int state, status;
1000         pid_t pid = task_pid_vnr(p);
1001         uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1002         struct waitid_info *infop;
1003
1004         if (!likely(wo->wo_flags & WEXITED))
1005                 return 0;
1006
1007         if (unlikely(wo->wo_flags & WNOWAIT)) {
1008                 status = p->exit_code;
1009                 get_task_struct(p);
1010                 read_unlock(&tasklist_lock);
1011                 sched_annotate_sleep();
1012                 if (wo->wo_rusage)
1013                         getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1014                 put_task_struct(p);
1015                 goto out_info;
1016         }
1017         /*
1018          * Move the task's state to DEAD/TRACE, only one thread can do this.
1019          */
1020         state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1021                 EXIT_TRACE : EXIT_DEAD;
1022         if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1023                 return 0;
1024         /*
1025          * We own this thread, nobody else can reap it.
1026          */
1027         read_unlock(&tasklist_lock);
1028         sched_annotate_sleep();
1029
1030         /*
1031          * Check thread_group_leader() to exclude the traced sub-threads.
1032          */
1033         if (state == EXIT_DEAD && thread_group_leader(p)) {
1034                 struct signal_struct *sig = p->signal;
1035                 struct signal_struct *psig = current->signal;
1036                 unsigned long maxrss;
1037                 u64 tgutime, tgstime;
1038
1039                 /*
1040                  * The resource counters for the group leader are in its
1041                  * own task_struct.  Those for dead threads in the group
1042                  * are in its signal_struct, as are those for the child
1043                  * processes it has previously reaped.  All these
1044                  * accumulate in the parent's signal_struct c* fields.
1045                  *
1046                  * We don't bother to take a lock here to protect these
1047                  * p->signal fields because the whole thread group is dead
1048                  * and nobody can change them.
1049                  *
1050                  * psig->stats_lock also protects us from our sub-theads
1051                  * which can reap other children at the same time. Until
1052                  * we change k_getrusage()-like users to rely on this lock
1053                  * we have to take ->siglock as well.
1054                  *
1055                  * We use thread_group_cputime_adjusted() to get times for
1056                  * the thread group, which consolidates times for all threads
1057                  * in the group including the group leader.
1058                  */
1059                 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1060                 spin_lock_irq(&current->sighand->siglock);
1061                 write_seqlock(&psig->stats_lock);
1062                 psig->cutime += tgutime + sig->cutime;
1063                 psig->cstime += tgstime + sig->cstime;
1064                 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1065                 psig->cmin_flt +=
1066                         p->min_flt + sig->min_flt + sig->cmin_flt;
1067                 psig->cmaj_flt +=
1068                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1069                 psig->cnvcsw +=
1070                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1071                 psig->cnivcsw +=
1072                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1073                 psig->cinblock +=
1074                         task_io_get_inblock(p) +
1075                         sig->inblock + sig->cinblock;
1076                 psig->coublock +=
1077                         task_io_get_oublock(p) +
1078                         sig->oublock + sig->coublock;
1079                 maxrss = max(sig->maxrss, sig->cmaxrss);
1080                 if (psig->cmaxrss < maxrss)
1081                         psig->cmaxrss = maxrss;
1082                 task_io_accounting_add(&psig->ioac, &p->ioac);
1083                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1084                 write_sequnlock(&psig->stats_lock);
1085                 spin_unlock_irq(&current->sighand->siglock);
1086         }
1087
1088         if (wo->wo_rusage)
1089                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1090         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1091                 ? p->signal->group_exit_code : p->exit_code;
1092         wo->wo_stat = status;
1093
1094         if (state == EXIT_TRACE) {
1095                 write_lock_irq(&tasklist_lock);
1096                 /* We dropped tasklist, ptracer could die and untrace */
1097                 ptrace_unlink(p);
1098
1099                 /* If parent wants a zombie, don't release it now */
1100                 state = EXIT_ZOMBIE;
1101                 if (do_notify_parent(p, p->exit_signal))
1102                         state = EXIT_DEAD;
1103                 p->exit_state = state;
1104                 write_unlock_irq(&tasklist_lock);
1105         }
1106         if (state == EXIT_DEAD)
1107                 release_task(p);
1108
1109 out_info:
1110         infop = wo->wo_info;
1111         if (infop) {
1112                 if ((status & 0x7f) == 0) {
1113                         infop->cause = CLD_EXITED;
1114                         infop->status = status >> 8;
1115                 } else {
1116                         infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1117                         infop->status = status & 0x7f;
1118                 }
1119                 infop->pid = pid;
1120                 infop->uid = uid;
1121         }
1122
1123         return pid;
1124 }
1125
1126 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1127 {
1128         if (ptrace) {
1129                 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1130                         return &p->exit_code;
1131         } else {
1132                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1133                         return &p->signal->group_exit_code;
1134         }
1135         return NULL;
1136 }
1137
1138 /**
1139  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1140  * @wo: wait options
1141  * @ptrace: is the wait for ptrace
1142  * @p: task to wait for
1143  *
1144  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1145  *
1146  * CONTEXT:
1147  * read_lock(&tasklist_lock), which is released if return value is
1148  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1149  *
1150  * RETURNS:
1151  * 0 if wait condition didn't exist and search for other wait conditions
1152  * should continue.  Non-zero return, -errno on failure and @p's pid on
1153  * success, implies that tasklist_lock is released and wait condition
1154  * search should terminate.
1155  */
1156 static int wait_task_stopped(struct wait_opts *wo,
1157                                 int ptrace, struct task_struct *p)
1158 {
1159         struct waitid_info *infop;
1160         int exit_code, *p_code, why;
1161         uid_t uid = 0; /* unneeded, required by compiler */
1162         pid_t pid;
1163
1164         /*
1165          * Traditionally we see ptrace'd stopped tasks regardless of options.
1166          */
1167         if (!ptrace && !(wo->wo_flags & WUNTRACED))
1168                 return 0;
1169
1170         if (!task_stopped_code(p, ptrace))
1171                 return 0;
1172
1173         exit_code = 0;
1174         spin_lock_irq(&p->sighand->siglock);
1175
1176         p_code = task_stopped_code(p, ptrace);
1177         if (unlikely(!p_code))
1178                 goto unlock_sig;
1179
1180         exit_code = *p_code;
1181         if (!exit_code)
1182                 goto unlock_sig;
1183
1184         if (!unlikely(wo->wo_flags & WNOWAIT))
1185                 *p_code = 0;
1186
1187         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1188 unlock_sig:
1189         spin_unlock_irq(&p->sighand->siglock);
1190         if (!exit_code)
1191                 return 0;
1192
1193         /*
1194          * Now we are pretty sure this task is interesting.
1195          * Make sure it doesn't get reaped out from under us while we
1196          * give up the lock and then examine it below.  We don't want to
1197          * keep holding onto the tasklist_lock while we call getrusage and
1198          * possibly take page faults for user memory.
1199          */
1200         get_task_struct(p);
1201         pid = task_pid_vnr(p);
1202         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1203         read_unlock(&tasklist_lock);
1204         sched_annotate_sleep();
1205         if (wo->wo_rusage)
1206                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1207         put_task_struct(p);
1208
1209         if (likely(!(wo->wo_flags & WNOWAIT)))
1210                 wo->wo_stat = (exit_code << 8) | 0x7f;
1211
1212         infop = wo->wo_info;
1213         if (infop) {
1214                 infop->cause = why;
1215                 infop->status = exit_code;
1216                 infop->pid = pid;
1217                 infop->uid = uid;
1218         }
1219         return pid;
1220 }
1221
1222 /*
1223  * Handle do_wait work for one task in a live, non-stopped state.
1224  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1225  * the lock and this task is uninteresting.  If we return nonzero, we have
1226  * released the lock and the system call should return.
1227  */
1228 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1229 {
1230         struct waitid_info *infop;
1231         pid_t pid;
1232         uid_t uid;
1233
1234         if (!unlikely(wo->wo_flags & WCONTINUED))
1235                 return 0;
1236
1237         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1238                 return 0;
1239
1240         spin_lock_irq(&p->sighand->siglock);
1241         /* Re-check with the lock held.  */
1242         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1243                 spin_unlock_irq(&p->sighand->siglock);
1244                 return 0;
1245         }
1246         if (!unlikely(wo->wo_flags & WNOWAIT))
1247                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1248         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1249         spin_unlock_irq(&p->sighand->siglock);
1250
1251         pid = task_pid_vnr(p);
1252         get_task_struct(p);
1253         read_unlock(&tasklist_lock);
1254         sched_annotate_sleep();
1255         if (wo->wo_rusage)
1256                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1257         put_task_struct(p);
1258
1259         infop = wo->wo_info;
1260         if (!infop) {
1261                 wo->wo_stat = 0xffff;
1262         } else {
1263                 infop->cause = CLD_CONTINUED;
1264                 infop->pid = pid;
1265                 infop->uid = uid;
1266                 infop->status = SIGCONT;
1267         }
1268         return pid;
1269 }
1270
1271 /*
1272  * Consider @p for a wait by @parent.
1273  *
1274  * -ECHILD should be in ->notask_error before the first call.
1275  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1276  * Returns zero if the search for a child should continue;
1277  * then ->notask_error is 0 if @p is an eligible child,
1278  * or still -ECHILD.
1279  */
1280 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1281                                 struct task_struct *p)
1282 {
1283         /*
1284          * We can race with wait_task_zombie() from another thread.
1285          * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1286          * can't confuse the checks below.
1287          */
1288         int exit_state = READ_ONCE(p->exit_state);
1289         int ret;
1290
1291         if (unlikely(exit_state == EXIT_DEAD))
1292                 return 0;
1293
1294         ret = eligible_child(wo, ptrace, p);
1295         if (!ret)
1296                 return ret;
1297
1298         if (unlikely(exit_state == EXIT_TRACE)) {
1299                 /*
1300                  * ptrace == 0 means we are the natural parent. In this case
1301                  * we should clear notask_error, debugger will notify us.
1302                  */
1303                 if (likely(!ptrace))
1304                         wo->notask_error = 0;
1305                 return 0;
1306         }
1307
1308         if (likely(!ptrace) && unlikely(p->ptrace)) {
1309                 /*
1310                  * If it is traced by its real parent's group, just pretend
1311                  * the caller is ptrace_do_wait() and reap this child if it
1312                  * is zombie.
1313                  *
1314                  * This also hides group stop state from real parent; otherwise
1315                  * a single stop can be reported twice as group and ptrace stop.
1316                  * If a ptracer wants to distinguish these two events for its
1317                  * own children it should create a separate process which takes
1318                  * the role of real parent.
1319                  */
1320                 if (!ptrace_reparented(p))
1321                         ptrace = 1;
1322         }
1323
1324         /* slay zombie? */
1325         if (exit_state == EXIT_ZOMBIE) {
1326                 /* we don't reap group leaders with subthreads */
1327                 if (!delay_group_leader(p)) {
1328                         /*
1329                          * A zombie ptracee is only visible to its ptracer.
1330                          * Notification and reaping will be cascaded to the
1331                          * real parent when the ptracer detaches.
1332                          */
1333                         if (unlikely(ptrace) || likely(!p->ptrace))
1334                                 return wait_task_zombie(wo, p);
1335                 }
1336
1337                 /*
1338                  * Allow access to stopped/continued state via zombie by
1339                  * falling through.  Clearing of notask_error is complex.
1340                  *
1341                  * When !@ptrace:
1342                  *
1343                  * If WEXITED is set, notask_error should naturally be
1344                  * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1345                  * so, if there are live subthreads, there are events to
1346                  * wait for.  If all subthreads are dead, it's still safe
1347                  * to clear - this function will be called again in finite
1348                  * amount time once all the subthreads are released and
1349                  * will then return without clearing.
1350                  *
1351                  * When @ptrace:
1352                  *
1353                  * Stopped state is per-task and thus can't change once the
1354                  * target task dies.  Only continued and exited can happen.
1355                  * Clear notask_error if WCONTINUED | WEXITED.
1356                  */
1357                 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1358                         wo->notask_error = 0;
1359         } else {
1360                 /*
1361                  * @p is alive and it's gonna stop, continue or exit, so
1362                  * there always is something to wait for.
1363                  */
1364                 wo->notask_error = 0;
1365         }
1366
1367         /*
1368          * Wait for stopped.  Depending on @ptrace, different stopped state
1369          * is used and the two don't interact with each other.
1370          */
1371         ret = wait_task_stopped(wo, ptrace, p);
1372         if (ret)
1373                 return ret;
1374
1375         /*
1376          * Wait for continued.  There's only one continued state and the
1377          * ptracer can consume it which can confuse the real parent.  Don't
1378          * use WCONTINUED from ptracer.  You don't need or want it.
1379          */
1380         return wait_task_continued(wo, p);
1381 }
1382
1383 /*
1384  * Do the work of do_wait() for one thread in the group, @tsk.
1385  *
1386  * -ECHILD should be in ->notask_error before the first call.
1387  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1388  * Returns zero if the search for a child should continue; then
1389  * ->notask_error is 0 if there were any eligible children,
1390  * or still -ECHILD.
1391  */
1392 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1393 {
1394         struct task_struct *p;
1395
1396         list_for_each_entry(p, &tsk->children, sibling) {
1397                 int ret = wait_consider_task(wo, 0, p);
1398
1399                 if (ret)
1400                         return ret;
1401         }
1402
1403         return 0;
1404 }
1405
1406 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1407 {
1408         struct task_struct *p;
1409
1410         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1411                 int ret = wait_consider_task(wo, 1, p);
1412
1413                 if (ret)
1414                         return ret;
1415         }
1416
1417         return 0;
1418 }
1419
1420 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1421                                 int sync, void *key)
1422 {
1423         struct wait_opts *wo = container_of(wait, struct wait_opts,
1424                                                 child_wait);
1425         struct task_struct *p = key;
1426
1427         if (!eligible_pid(wo, p))
1428                 return 0;
1429
1430         if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1431                 return 0;
1432
1433         return default_wake_function(wait, mode, sync, key);
1434 }
1435
1436 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1437 {
1438         __wake_up_sync_key(&parent->signal->wait_chldexit,
1439                            TASK_INTERRUPTIBLE, p);
1440 }
1441
1442 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1443                                  struct task_struct *target)
1444 {
1445         struct task_struct *parent =
1446                 !ptrace ? target->real_parent : target->parent;
1447
1448         return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1449                                      same_thread_group(current, parent));
1450 }
1451
1452 /*
1453  * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1454  * and tracee lists to find the target task.
1455  */
1456 static int do_wait_pid(struct wait_opts *wo)
1457 {
1458         bool ptrace;
1459         struct task_struct *target;
1460         int retval;
1461
1462         ptrace = false;
1463         target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1464         if (target && is_effectively_child(wo, ptrace, target)) {
1465                 retval = wait_consider_task(wo, ptrace, target);
1466                 if (retval)
1467                         return retval;
1468         }
1469
1470         ptrace = true;
1471         target = pid_task(wo->wo_pid, PIDTYPE_PID);
1472         if (target && target->ptrace &&
1473             is_effectively_child(wo, ptrace, target)) {
1474                 retval = wait_consider_task(wo, ptrace, target);
1475                 if (retval)
1476                         return retval;
1477         }
1478
1479         return 0;
1480 }
1481
1482 static long do_wait(struct wait_opts *wo)
1483 {
1484         int retval;
1485
1486         trace_sched_process_wait(wo->wo_pid);
1487
1488         init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1489         wo->child_wait.private = current;
1490         add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1491 repeat:
1492         /*
1493          * If there is nothing that can match our criteria, just get out.
1494          * We will clear ->notask_error to zero if we see any child that
1495          * might later match our criteria, even if we are not able to reap
1496          * it yet.
1497          */
1498         wo->notask_error = -ECHILD;
1499         if ((wo->wo_type < PIDTYPE_MAX) &&
1500            (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1501                 goto notask;
1502
1503         set_current_state(TASK_INTERRUPTIBLE);
1504         read_lock(&tasklist_lock);
1505
1506         if (wo->wo_type == PIDTYPE_PID) {
1507                 retval = do_wait_pid(wo);
1508                 if (retval)
1509                         goto end;
1510         } else {
1511                 struct task_struct *tsk = current;
1512
1513                 do {
1514                         retval = do_wait_thread(wo, tsk);
1515                         if (retval)
1516                                 goto end;
1517
1518                         retval = ptrace_do_wait(wo, tsk);
1519                         if (retval)
1520                                 goto end;
1521
1522                         if (wo->wo_flags & __WNOTHREAD)
1523                                 break;
1524                 } while_each_thread(current, tsk);
1525         }
1526         read_unlock(&tasklist_lock);
1527
1528 notask:
1529         retval = wo->notask_error;
1530         if (!retval && !(wo->wo_flags & WNOHANG)) {
1531                 retval = -ERESTARTSYS;
1532                 if (!signal_pending(current)) {
1533                         schedule();
1534                         goto repeat;
1535                 }
1536         }
1537 end:
1538         __set_current_state(TASK_RUNNING);
1539         remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1540         return retval;
1541 }
1542
1543 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1544                           int options, struct rusage *ru)
1545 {
1546         struct wait_opts wo;
1547         struct pid *pid = NULL;
1548         enum pid_type type;
1549         long ret;
1550         unsigned int f_flags = 0;
1551
1552         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1553                         __WNOTHREAD|__WCLONE|__WALL))
1554                 return -EINVAL;
1555         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1556                 return -EINVAL;
1557
1558         switch (which) {
1559         case P_ALL:
1560                 type = PIDTYPE_MAX;
1561                 break;
1562         case P_PID:
1563                 type = PIDTYPE_PID;
1564                 if (upid <= 0)
1565                         return -EINVAL;
1566
1567                 pid = find_get_pid(upid);
1568                 break;
1569         case P_PGID:
1570                 type = PIDTYPE_PGID;
1571                 if (upid < 0)
1572                         return -EINVAL;
1573
1574                 if (upid)
1575                         pid = find_get_pid(upid);
1576                 else
1577                         pid = get_task_pid(current, PIDTYPE_PGID);
1578                 break;
1579         case P_PIDFD:
1580                 type = PIDTYPE_PID;
1581                 if (upid < 0)
1582                         return -EINVAL;
1583
1584                 pid = pidfd_get_pid(upid, &f_flags);
1585                 if (IS_ERR(pid))
1586                         return PTR_ERR(pid);
1587
1588                 break;
1589         default:
1590                 return -EINVAL;
1591         }
1592
1593         wo.wo_type      = type;
1594         wo.wo_pid       = pid;
1595         wo.wo_flags     = options;
1596         wo.wo_info      = infop;
1597         wo.wo_rusage    = ru;
1598         if (f_flags & O_NONBLOCK)
1599                 wo.wo_flags |= WNOHANG;
1600
1601         ret = do_wait(&wo);
1602         if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1603                 ret = -EAGAIN;
1604
1605         put_pid(pid);
1606         return ret;
1607 }
1608
1609 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1610                 infop, int, options, struct rusage __user *, ru)
1611 {
1612         struct rusage r;
1613         struct waitid_info info = {.status = 0};
1614         long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1615         int signo = 0;
1616
1617         if (err > 0) {
1618                 signo = SIGCHLD;
1619                 err = 0;
1620                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1621                         return -EFAULT;
1622         }
1623         if (!infop)
1624                 return err;
1625
1626         if (!user_write_access_begin(infop, sizeof(*infop)))
1627                 return -EFAULT;
1628
1629         unsafe_put_user(signo, &infop->si_signo, Efault);
1630         unsafe_put_user(0, &infop->si_errno, Efault);
1631         unsafe_put_user(info.cause, &infop->si_code, Efault);
1632         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1633         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1634         unsafe_put_user(info.status, &infop->si_status, Efault);
1635         user_write_access_end();
1636         return err;
1637 Efault:
1638         user_write_access_end();
1639         return -EFAULT;
1640 }
1641
1642 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1643                   struct rusage *ru)
1644 {
1645         struct wait_opts wo;
1646         struct pid *pid = NULL;
1647         enum pid_type type;
1648         long ret;
1649
1650         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1651                         __WNOTHREAD|__WCLONE|__WALL))
1652                 return -EINVAL;
1653
1654         /* -INT_MIN is not defined */
1655         if (upid == INT_MIN)
1656                 return -ESRCH;
1657
1658         if (upid == -1)
1659                 type = PIDTYPE_MAX;
1660         else if (upid < 0) {
1661                 type = PIDTYPE_PGID;
1662                 pid = find_get_pid(-upid);
1663         } else if (upid == 0) {
1664                 type = PIDTYPE_PGID;
1665                 pid = get_task_pid(current, PIDTYPE_PGID);
1666         } else /* upid > 0 */ {
1667                 type = PIDTYPE_PID;
1668                 pid = find_get_pid(upid);
1669         }
1670
1671         wo.wo_type      = type;
1672         wo.wo_pid       = pid;
1673         wo.wo_flags     = options | WEXITED;
1674         wo.wo_info      = NULL;
1675         wo.wo_stat      = 0;
1676         wo.wo_rusage    = ru;
1677         ret = do_wait(&wo);
1678         put_pid(pid);
1679         if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1680                 ret = -EFAULT;
1681
1682         return ret;
1683 }
1684
1685 int kernel_wait(pid_t pid, int *stat)
1686 {
1687         struct wait_opts wo = {
1688                 .wo_type        = PIDTYPE_PID,
1689                 .wo_pid         = find_get_pid(pid),
1690                 .wo_flags       = WEXITED,
1691         };
1692         int ret;
1693
1694         ret = do_wait(&wo);
1695         if (ret > 0 && wo.wo_stat)
1696                 *stat = wo.wo_stat;
1697         put_pid(wo.wo_pid);
1698         return ret;
1699 }
1700
1701 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1702                 int, options, struct rusage __user *, ru)
1703 {
1704         struct rusage r;
1705         long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1706
1707         if (err > 0) {
1708                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1709                         return -EFAULT;
1710         }
1711         return err;
1712 }
1713
1714 #ifdef __ARCH_WANT_SYS_WAITPID
1715
1716 /*
1717  * sys_waitpid() remains for compatibility. waitpid() should be
1718  * implemented by calling sys_wait4() from libc.a.
1719  */
1720 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1721 {
1722         return kernel_wait4(pid, stat_addr, options, NULL);
1723 }
1724
1725 #endif
1726
1727 #ifdef CONFIG_COMPAT
1728 COMPAT_SYSCALL_DEFINE4(wait4,
1729         compat_pid_t, pid,
1730         compat_uint_t __user *, stat_addr,
1731         int, options,
1732         struct compat_rusage __user *, ru)
1733 {
1734         struct rusage r;
1735         long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1736         if (err > 0) {
1737                 if (ru && put_compat_rusage(&r, ru))
1738                         return -EFAULT;
1739         }
1740         return err;
1741 }
1742
1743 COMPAT_SYSCALL_DEFINE5(waitid,
1744                 int, which, compat_pid_t, pid,
1745                 struct compat_siginfo __user *, infop, int, options,
1746                 struct compat_rusage __user *, uru)
1747 {
1748         struct rusage ru;
1749         struct waitid_info info = {.status = 0};
1750         long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1751         int signo = 0;
1752         if (err > 0) {
1753                 signo = SIGCHLD;
1754                 err = 0;
1755                 if (uru) {
1756                         /* kernel_waitid() overwrites everything in ru */
1757                         if (COMPAT_USE_64BIT_TIME)
1758                                 err = copy_to_user(uru, &ru, sizeof(ru));
1759                         else
1760                                 err = put_compat_rusage(&ru, uru);
1761                         if (err)
1762                                 return -EFAULT;
1763                 }
1764         }
1765
1766         if (!infop)
1767                 return err;
1768
1769         if (!user_write_access_begin(infop, sizeof(*infop)))
1770                 return -EFAULT;
1771
1772         unsafe_put_user(signo, &infop->si_signo, Efault);
1773         unsafe_put_user(0, &infop->si_errno, Efault);
1774         unsafe_put_user(info.cause, &infop->si_code, Efault);
1775         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1776         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1777         unsafe_put_user(info.status, &infop->si_status, Efault);
1778         user_write_access_end();
1779         return err;
1780 Efault:
1781         user_write_access_end();
1782         return -EFAULT;
1783 }
1784 #endif
1785
1786 /**
1787  * thread_group_exited - check that a thread group has exited
1788  * @pid: tgid of thread group to be checked.
1789  *
1790  * Test if the thread group represented by tgid has exited (all
1791  * threads are zombies, dead or completely gone).
1792  *
1793  * Return: true if the thread group has exited. false otherwise.
1794  */
1795 bool thread_group_exited(struct pid *pid)
1796 {
1797         struct task_struct *task;
1798         bool exited;
1799
1800         rcu_read_lock();
1801         task = pid_task(pid, PIDTYPE_PID);
1802         exited = !task ||
1803                 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1804         rcu_read_unlock();
1805
1806         return exited;
1807 }
1808 EXPORT_SYMBOL(thread_group_exited);
1809
1810 __weak void abort(void)
1811 {
1812         BUG();
1813
1814         /* if that doesn't kill us, halt */
1815         panic("Oops failed to kill thread");
1816 }
1817 EXPORT_SYMBOL(abort);