s390/mcck: fix calculation of SIE critical section size
[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 #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         exit_task_sigqueue_cache(tsk);
166 }
167
168 static void delayed_put_task_struct(struct rcu_head *rhp)
169 {
170         struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
171
172         perf_event_delayed_put(tsk);
173         trace_sched_process_free(tsk);
174         put_task_struct(tsk);
175 }
176
177 void put_task_struct_rcu_user(struct task_struct *task)
178 {
179         if (refcount_dec_and_test(&task->rcu_users))
180                 call_rcu(&task->rcu, delayed_put_task_struct);
181 }
182
183 void release_task(struct task_struct *p)
184 {
185         struct task_struct *leader;
186         struct pid *thread_pid;
187         int zap_leader;
188 repeat:
189         /* don't need to get the RCU readlock here - the process is dead and
190          * can't be modifying its own credentials. But shut RCU-lockdep up */
191         rcu_read_lock();
192         atomic_dec(&__task_cred(p)->user->processes);
193         rcu_read_unlock();
194
195         cgroup_release(p);
196
197         write_lock_irq(&tasklist_lock);
198         ptrace_release_task(p);
199         thread_pid = get_pid(p->thread_pid);
200         __exit_signal(p);
201
202         /*
203          * If we are the last non-leader member of the thread
204          * group, and the leader is zombie, then notify the
205          * group leader's parent process. (if it wants notification.)
206          */
207         zap_leader = 0;
208         leader = p->group_leader;
209         if (leader != p && thread_group_empty(leader)
210                         && leader->exit_state == EXIT_ZOMBIE) {
211                 /*
212                  * If we were the last child thread and the leader has
213                  * exited already, and the leader's parent ignores SIGCHLD,
214                  * then we are the one who should release the leader.
215                  */
216                 zap_leader = do_notify_parent(leader, leader->exit_signal);
217                 if (zap_leader)
218                         leader->exit_state = EXIT_DEAD;
219         }
220
221         write_unlock_irq(&tasklist_lock);
222         seccomp_filter_release(p);
223         proc_flush_pid(thread_pid);
224         put_pid(thread_pid);
225         release_thread(p);
226         put_task_struct_rcu_user(p);
227
228         p = leader;
229         if (unlikely(zap_leader))
230                 goto repeat;
231 }
232
233 int rcuwait_wake_up(struct rcuwait *w)
234 {
235         int ret = 0;
236         struct task_struct *task;
237
238         rcu_read_lock();
239
240         /*
241          * Order condition vs @task, such that everything prior to the load
242          * of @task is visible. This is the condition as to why the user called
243          * rcuwait_wake() in the first place. Pairs with set_current_state()
244          * barrier (A) in rcuwait_wait_event().
245          *
246          *    WAIT                WAKE
247          *    [S] tsk = current   [S] cond = true
248          *        MB (A)              MB (B)
249          *    [L] cond            [L] tsk
250          */
251         smp_mb(); /* (B) */
252
253         task = rcu_dereference(w->task);
254         if (task)
255                 ret = wake_up_process(task);
256         rcu_read_unlock();
257
258         return ret;
259 }
260 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
261
262 /*
263  * Determine if a process group is "orphaned", according to the POSIX
264  * definition in 2.2.2.52.  Orphaned process groups are not to be affected
265  * by terminal-generated stop signals.  Newly orphaned process groups are
266  * to receive a SIGHUP and a SIGCONT.
267  *
268  * "I ask you, have you ever known what it is to be an orphan?"
269  */
270 static int will_become_orphaned_pgrp(struct pid *pgrp,
271                                         struct task_struct *ignored_task)
272 {
273         struct task_struct *p;
274
275         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
276                 if ((p == ignored_task) ||
277                     (p->exit_state && thread_group_empty(p)) ||
278                     is_global_init(p->real_parent))
279                         continue;
280
281                 if (task_pgrp(p->real_parent) != pgrp &&
282                     task_session(p->real_parent) == task_session(p))
283                         return 0;
284         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
285
286         return 1;
287 }
288
289 int is_current_pgrp_orphaned(void)
290 {
291         int retval;
292
293         read_lock(&tasklist_lock);
294         retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
295         read_unlock(&tasklist_lock);
296
297         return retval;
298 }
299
300 static bool has_stopped_jobs(struct pid *pgrp)
301 {
302         struct task_struct *p;
303
304         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
305                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
306                         return true;
307         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
308
309         return false;
310 }
311
312 /*
313  * Check to see if any process groups have become orphaned as
314  * a result of our exiting, and if they have any stopped jobs,
315  * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
316  */
317 static void
318 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
319 {
320         struct pid *pgrp = task_pgrp(tsk);
321         struct task_struct *ignored_task = tsk;
322
323         if (!parent)
324                 /* exit: our father is in a different pgrp than
325                  * we are and we were the only connection outside.
326                  */
327                 parent = tsk->real_parent;
328         else
329                 /* reparent: our child is in a different pgrp than
330                  * we are, and it was the only connection outside.
331                  */
332                 ignored_task = NULL;
333
334         if (task_pgrp(parent) != pgrp &&
335             task_session(parent) == task_session(tsk) &&
336             will_become_orphaned_pgrp(pgrp, ignored_task) &&
337             has_stopped_jobs(pgrp)) {
338                 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
339                 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
340         }
341 }
342
343 #ifdef CONFIG_MEMCG
344 /*
345  * A task is exiting.   If it owned this mm, find a new owner for the mm.
346  */
347 void mm_update_next_owner(struct mm_struct *mm)
348 {
349         struct task_struct *c, *g, *p = current;
350
351 retry:
352         /*
353          * If the exiting or execing task is not the owner, it's
354          * someone else's problem.
355          */
356         if (mm->owner != p)
357                 return;
358         /*
359          * The current owner is exiting/execing and there are no other
360          * candidates.  Do not leave the mm pointing to a possibly
361          * freed task structure.
362          */
363         if (atomic_read(&mm->mm_users) <= 1) {
364                 WRITE_ONCE(mm->owner, NULL);
365                 return;
366         }
367
368         read_lock(&tasklist_lock);
369         /*
370          * Search in the children
371          */
372         list_for_each_entry(c, &p->children, sibling) {
373                 if (c->mm == mm)
374                         goto assign_new_owner;
375         }
376
377         /*
378          * Search in the siblings
379          */
380         list_for_each_entry(c, &p->real_parent->children, sibling) {
381                 if (c->mm == mm)
382                         goto assign_new_owner;
383         }
384
385         /*
386          * Search through everything else, we should not get here often.
387          */
388         for_each_process(g) {
389                 if (g->flags & PF_KTHREAD)
390                         continue;
391                 for_each_thread(g, c) {
392                         if (c->mm == mm)
393                                 goto assign_new_owner;
394                         if (c->mm)
395                                 break;
396                 }
397         }
398         read_unlock(&tasklist_lock);
399         /*
400          * We found no owner yet mm_users > 1: this implies that we are
401          * most likely racing with swapoff (try_to_unuse()) or /proc or
402          * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
403          */
404         WRITE_ONCE(mm->owner, NULL);
405         return;
406
407 assign_new_owner:
408         BUG_ON(c == p);
409         get_task_struct(c);
410         /*
411          * The task_lock protects c->mm from changing.
412          * We always want mm->owner->mm == mm
413          */
414         task_lock(c);
415         /*
416          * Delay read_unlock() till we have the task_lock()
417          * to ensure that c does not slip away underneath us
418          */
419         read_unlock(&tasklist_lock);
420         if (c->mm != mm) {
421                 task_unlock(c);
422                 put_task_struct(c);
423                 goto retry;
424         }
425         WRITE_ONCE(mm->owner, c);
426         task_unlock(c);
427         put_task_struct(c);
428 }
429 #endif /* CONFIG_MEMCG */
430
431 /*
432  * Turn us into a lazy TLB process if we
433  * aren't already..
434  */
435 static void exit_mm(void)
436 {
437         struct mm_struct *mm = current->mm;
438         struct core_state *core_state;
439
440         exit_mm_release(current, mm);
441         if (!mm)
442                 return;
443         sync_mm_rss(mm);
444         /*
445          * Serialize with any possible pending coredump.
446          * We must hold mmap_lock around checking core_state
447          * and clearing tsk->mm.  The core-inducing thread
448          * will increment ->nr_threads for each thread in the
449          * group with ->mm != NULL.
450          */
451         mmap_read_lock(mm);
452         core_state = mm->core_state;
453         if (core_state) {
454                 struct core_thread self;
455
456                 mmap_read_unlock(mm);
457
458                 self.task = current;
459                 if (self.task->flags & PF_SIGNALED)
460                         self.next = xchg(&core_state->dumper.next, &self);
461                 else
462                         self.task = NULL;
463                 /*
464                  * Implies mb(), the result of xchg() must be visible
465                  * to core_state->dumper.
466                  */
467                 if (atomic_dec_and_test(&core_state->nr_threads))
468                         complete(&core_state->startup);
469
470                 for (;;) {
471                         set_current_state(TASK_UNINTERRUPTIBLE);
472                         if (!self.task) /* see coredump_finish() */
473                                 break;
474                         freezable_schedule();
475                 }
476                 __set_current_state(TASK_RUNNING);
477                 mmap_read_lock(mm);
478         }
479         mmgrab(mm);
480         BUG_ON(mm != current->active_mm);
481         /* more a memory barrier than a real lock */
482         task_lock(current);
483         /*
484          * When a thread stops operating on an address space, the loop
485          * in membarrier_private_expedited() may not observe that
486          * tsk->mm, and the loop in membarrier_global_expedited() may
487          * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
488          * rq->membarrier_state, so those would not issue an IPI.
489          * Membarrier requires a memory barrier after accessing
490          * user-space memory, before clearing tsk->mm or the
491          * rq->membarrier_state.
492          */
493         smp_mb__after_spinlock();
494         local_irq_disable();
495         current->mm = NULL;
496         membarrier_update_current_mm(NULL);
497         enter_lazy_tlb(mm, current);
498         local_irq_enable();
499         task_unlock(current);
500         mmap_read_unlock(mm);
501         mm_update_next_owner(mm);
502         mmput(mm);
503         if (test_thread_flag(TIF_MEMDIE))
504                 exit_oom_victim();
505 }
506
507 static struct task_struct *find_alive_thread(struct task_struct *p)
508 {
509         struct task_struct *t;
510
511         for_each_thread(p, t) {
512                 if (!(t->flags & PF_EXITING))
513                         return t;
514         }
515         return NULL;
516 }
517
518 static struct task_struct *find_child_reaper(struct task_struct *father,
519                                                 struct list_head *dead)
520         __releases(&tasklist_lock)
521         __acquires(&tasklist_lock)
522 {
523         struct pid_namespace *pid_ns = task_active_pid_ns(father);
524         struct task_struct *reaper = pid_ns->child_reaper;
525         struct task_struct *p, *n;
526
527         if (likely(reaper != father))
528                 return reaper;
529
530         reaper = find_alive_thread(father);
531         if (reaper) {
532                 pid_ns->child_reaper = reaper;
533                 return reaper;
534         }
535
536         write_unlock_irq(&tasklist_lock);
537
538         list_for_each_entry_safe(p, n, dead, ptrace_entry) {
539                 list_del_init(&p->ptrace_entry);
540                 release_task(p);
541         }
542
543         zap_pid_ns_processes(pid_ns);
544         write_lock_irq(&tasklist_lock);
545
546         return father;
547 }
548
549 /*
550  * When we die, we re-parent all our children, and try to:
551  * 1. give them to another thread in our thread group, if such a member exists
552  * 2. give it to the first ancestor process which prctl'd itself as a
553  *    child_subreaper for its children (like a service manager)
554  * 3. give it to the init process (PID 1) in our pid namespace
555  */
556 static struct task_struct *find_new_reaper(struct task_struct *father,
557                                            struct task_struct *child_reaper)
558 {
559         struct task_struct *thread, *reaper;
560
561         thread = find_alive_thread(father);
562         if (thread)
563                 return thread;
564
565         if (father->signal->has_child_subreaper) {
566                 unsigned int ns_level = task_pid(father)->level;
567                 /*
568                  * Find the first ->is_child_subreaper ancestor in our pid_ns.
569                  * We can't check reaper != child_reaper to ensure we do not
570                  * cross the namespaces, the exiting parent could be injected
571                  * by setns() + fork().
572                  * We check pid->level, this is slightly more efficient than
573                  * task_active_pid_ns(reaper) != task_active_pid_ns(father).
574                  */
575                 for (reaper = father->real_parent;
576                      task_pid(reaper)->level == ns_level;
577                      reaper = reaper->real_parent) {
578                         if (reaper == &init_task)
579                                 break;
580                         if (!reaper->signal->is_child_subreaper)
581                                 continue;
582                         thread = find_alive_thread(reaper);
583                         if (thread)
584                                 return thread;
585                 }
586         }
587
588         return child_reaper;
589 }
590
591 /*
592 * Any that need to be release_task'd are put on the @dead list.
593  */
594 static void reparent_leader(struct task_struct *father, struct task_struct *p,
595                                 struct list_head *dead)
596 {
597         if (unlikely(p->exit_state == EXIT_DEAD))
598                 return;
599
600         /* We don't want people slaying init. */
601         p->exit_signal = SIGCHLD;
602
603         /* If it has exited notify the new parent about this child's death. */
604         if (!p->ptrace &&
605             p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
606                 if (do_notify_parent(p, p->exit_signal)) {
607                         p->exit_state = EXIT_DEAD;
608                         list_add(&p->ptrace_entry, dead);
609                 }
610         }
611
612         kill_orphaned_pgrp(p, father);
613 }
614
615 /*
616  * This does two things:
617  *
618  * A.  Make init inherit all the child processes
619  * B.  Check to see if any process groups have become orphaned
620  *      as a result of our exiting, and if they have any stopped
621  *      jobs, send them a SIGHUP and then a SIGCONT.  (POSIX 3.2.2.2)
622  */
623 static void forget_original_parent(struct task_struct *father,
624                                         struct list_head *dead)
625 {
626         struct task_struct *p, *t, *reaper;
627
628         if (unlikely(!list_empty(&father->ptraced)))
629                 exit_ptrace(father, dead);
630
631         /* Can drop and reacquire tasklist_lock */
632         reaper = find_child_reaper(father, dead);
633         if (list_empty(&father->children))
634                 return;
635
636         reaper = find_new_reaper(father, reaper);
637         list_for_each_entry(p, &father->children, sibling) {
638                 for_each_thread(p, t) {
639                         RCU_INIT_POINTER(t->real_parent, reaper);
640                         BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
641                         if (likely(!t->ptrace))
642                                 t->parent = t->real_parent;
643                         if (t->pdeath_signal)
644                                 group_send_sig_info(t->pdeath_signal,
645                                                     SEND_SIG_NOINFO, t,
646                                                     PIDTYPE_TGID);
647                 }
648                 /*
649                  * If this is a threaded reparent there is no need to
650                  * notify anyone anything has happened.
651                  */
652                 if (!same_thread_group(reaper, father))
653                         reparent_leader(father, p, dead);
654         }
655         list_splice_tail_init(&father->children, &reaper->children);
656 }
657
658 /*
659  * Send signals to all our closest relatives so that they know
660  * to properly mourn us..
661  */
662 static void exit_notify(struct task_struct *tsk, int group_dead)
663 {
664         bool autoreap;
665         struct task_struct *p, *n;
666         LIST_HEAD(dead);
667
668         write_lock_irq(&tasklist_lock);
669         forget_original_parent(tsk, &dead);
670
671         if (group_dead)
672                 kill_orphaned_pgrp(tsk->group_leader, NULL);
673
674         tsk->exit_state = EXIT_ZOMBIE;
675         if (unlikely(tsk->ptrace)) {
676                 int sig = thread_group_leader(tsk) &&
677                                 thread_group_empty(tsk) &&
678                                 !ptrace_reparented(tsk) ?
679                         tsk->exit_signal : SIGCHLD;
680                 autoreap = do_notify_parent(tsk, sig);
681         } else if (thread_group_leader(tsk)) {
682                 autoreap = thread_group_empty(tsk) &&
683                         do_notify_parent(tsk, tsk->exit_signal);
684         } else {
685                 autoreap = true;
686         }
687
688         if (autoreap) {
689                 tsk->exit_state = EXIT_DEAD;
690                 list_add(&tsk->ptrace_entry, &dead);
691         }
692
693         /* mt-exec, de_thread() is waiting for group leader */
694         if (unlikely(tsk->signal->notify_count < 0))
695                 wake_up_process(tsk->signal->group_exit_task);
696         write_unlock_irq(&tasklist_lock);
697
698         list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
699                 list_del_init(&p->ptrace_entry);
700                 release_task(p);
701         }
702 }
703
704 #ifdef CONFIG_DEBUG_STACK_USAGE
705 static void check_stack_usage(void)
706 {
707         static DEFINE_SPINLOCK(low_water_lock);
708         static int lowest_to_date = THREAD_SIZE;
709         unsigned long free;
710
711         free = stack_not_used(current);
712
713         if (free >= lowest_to_date)
714                 return;
715
716         spin_lock(&low_water_lock);
717         if (free < lowest_to_date) {
718                 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
719                         current->comm, task_pid_nr(current), free);
720                 lowest_to_date = free;
721         }
722         spin_unlock(&low_water_lock);
723 }
724 #else
725 static inline void check_stack_usage(void) {}
726 #endif
727
728 void __noreturn do_exit(long code)
729 {
730         struct task_struct *tsk = current;
731         int group_dead;
732
733         /*
734          * We can get here from a kernel oops, sometimes with preemption off.
735          * Start by checking for critical errors.
736          * Then fix up important state like USER_DS and preemption.
737          * Then do everything else.
738          */
739
740         WARN_ON(blk_needs_flush_plug(tsk));
741
742         if (unlikely(in_interrupt()))
743                 panic("Aiee, killing interrupt handler!");
744         if (unlikely(!tsk->pid))
745                 panic("Attempted to kill the idle task!");
746
747         /*
748          * If do_exit is called because this processes oopsed, it's possible
749          * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
750          * continuing. Amongst other possible reasons, this is to prevent
751          * mm_release()->clear_child_tid() from writing to a user-controlled
752          * kernel address.
753          */
754         force_uaccess_begin();
755
756         if (unlikely(in_atomic())) {
757                 pr_info("note: %s[%d] exited with preempt_count %d\n",
758                         current->comm, task_pid_nr(current),
759                         preempt_count());
760                 preempt_count_set(PREEMPT_ENABLED);
761         }
762
763         profile_task_exit(tsk);
764         kcov_task_exit(tsk);
765
766         ptrace_event(PTRACE_EVENT_EXIT, code);
767
768         validate_creds_for_do_exit(tsk);
769
770         /*
771          * We're taking recursive faults here in do_exit. Safest is to just
772          * leave this task alone and wait for reboot.
773          */
774         if (unlikely(tsk->flags & PF_EXITING)) {
775                 pr_alert("Fixing recursive fault but reboot is needed!\n");
776                 futex_exit_recursive(tsk);
777                 set_current_state(TASK_UNINTERRUPTIBLE);
778                 schedule();
779         }
780
781         io_uring_files_cancel(tsk->files);
782         exit_signals(tsk);  /* sets PF_EXITING */
783
784         /* sync mm's RSS info before statistics gathering */
785         if (tsk->mm)
786                 sync_mm_rss(tsk->mm);
787         acct_update_integrals(tsk);
788         group_dead = atomic_dec_and_test(&tsk->signal->live);
789         if (group_dead) {
790                 /*
791                  * If the last thread of global init has exited, panic
792                  * immediately to get a useable coredump.
793                  */
794                 if (unlikely(is_global_init(tsk)))
795                         panic("Attempted to kill init! exitcode=0x%08x\n",
796                                 tsk->signal->group_exit_code ?: (int)code);
797
798 #ifdef CONFIG_POSIX_TIMERS
799                 hrtimer_cancel(&tsk->signal->real_timer);
800                 exit_itimers(tsk->signal);
801 #endif
802                 if (tsk->mm)
803                         setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
804         }
805         acct_collect(code, group_dead);
806         if (group_dead)
807                 tty_audit_exit();
808         audit_free(tsk);
809
810         tsk->exit_code = code;
811         taskstats_exit(tsk, group_dead);
812
813         exit_mm();
814
815         if (group_dead)
816                 acct_process();
817         trace_sched_process_exit(tsk);
818
819         exit_sem(tsk);
820         exit_shm(tsk);
821         exit_files(tsk);
822         exit_fs(tsk);
823         if (group_dead)
824                 disassociate_ctty(1);
825         exit_task_namespaces(tsk);
826         exit_task_work(tsk);
827         exit_thread(tsk);
828
829         /*
830          * Flush inherited counters to the parent - before the parent
831          * gets woken up by child-exit notifications.
832          *
833          * because of cgroup mode, must be called before cgroup_exit()
834          */
835         perf_event_exit_task(tsk);
836
837         sched_autogroup_exit_task(tsk);
838         cgroup_exit(tsk);
839
840         /*
841          * FIXME: do that only when needed, using sched_exit tracepoint
842          */
843         flush_ptrace_hw_breakpoint(tsk);
844
845         exit_tasks_rcu_start();
846         exit_notify(tsk, group_dead);
847         proc_exit_connector(tsk);
848         mpol_put_task_policy(tsk);
849 #ifdef CONFIG_FUTEX
850         if (unlikely(current->pi_state_cache))
851                 kfree(current->pi_state_cache);
852 #endif
853         /*
854          * Make sure we are holding no locks:
855          */
856         debug_check_no_locks_held();
857
858         if (tsk->io_context)
859                 exit_io_context(tsk);
860
861         if (tsk->splice_pipe)
862                 free_pipe_info(tsk->splice_pipe);
863
864         if (tsk->task_frag.page)
865                 put_page(tsk->task_frag.page);
866
867         validate_creds_for_do_exit(tsk);
868
869         check_stack_usage();
870         preempt_disable();
871         if (tsk->nr_dirtied)
872                 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
873         exit_rcu();
874         exit_tasks_rcu_finish();
875
876         lockdep_free_task(tsk);
877         do_task_dead();
878 }
879 EXPORT_SYMBOL_GPL(do_exit);
880
881 void complete_and_exit(struct completion *comp, long code)
882 {
883         if (comp)
884                 complete(comp);
885
886         do_exit(code);
887 }
888 EXPORT_SYMBOL(complete_and_exit);
889
890 SYSCALL_DEFINE1(exit, int, error_code)
891 {
892         do_exit((error_code&0xff)<<8);
893 }
894
895 /*
896  * Take down every thread in the group.  This is called by fatal signals
897  * as well as by sys_exit_group (below).
898  */
899 void
900 do_group_exit(int exit_code)
901 {
902         struct signal_struct *sig = current->signal;
903
904         BUG_ON(exit_code & 0x80); /* core dumps don't get here */
905
906         if (signal_group_exit(sig))
907                 exit_code = sig->group_exit_code;
908         else if (!thread_group_empty(current)) {
909                 struct sighand_struct *const sighand = current->sighand;
910
911                 spin_lock_irq(&sighand->siglock);
912                 if (signal_group_exit(sig))
913                         /* Another thread got here before we took the lock.  */
914                         exit_code = sig->group_exit_code;
915                 else {
916                         sig->group_exit_code = exit_code;
917                         sig->flags = SIGNAL_GROUP_EXIT;
918                         zap_other_threads(current);
919                 }
920                 spin_unlock_irq(&sighand->siglock);
921         }
922
923         do_exit(exit_code);
924         /* NOTREACHED */
925 }
926
927 /*
928  * this kills every thread in the thread group. Note that any externally
929  * wait4()-ing process will get the correct exit code - even if this
930  * thread is not the thread group leader.
931  */
932 SYSCALL_DEFINE1(exit_group, int, error_code)
933 {
934         do_group_exit((error_code & 0xff) << 8);
935         /* NOTREACHED */
936         return 0;
937 }
938
939 struct waitid_info {
940         pid_t pid;
941         uid_t uid;
942         int status;
943         int cause;
944 };
945
946 struct wait_opts {
947         enum pid_type           wo_type;
948         int                     wo_flags;
949         struct pid              *wo_pid;
950
951         struct waitid_info      *wo_info;
952         int                     wo_stat;
953         struct rusage           *wo_rusage;
954
955         wait_queue_entry_t              child_wait;
956         int                     notask_error;
957 };
958
959 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
960 {
961         return  wo->wo_type == PIDTYPE_MAX ||
962                 task_pid_type(p, wo->wo_type) == wo->wo_pid;
963 }
964
965 static int
966 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
967 {
968         if (!eligible_pid(wo, p))
969                 return 0;
970
971         /*
972          * Wait for all children (clone and not) if __WALL is set or
973          * if it is traced by us.
974          */
975         if (ptrace || (wo->wo_flags & __WALL))
976                 return 1;
977
978         /*
979          * Otherwise, wait for clone children *only* if __WCLONE is set;
980          * otherwise, wait for non-clone children *only*.
981          *
982          * Note: a "clone" child here is one that reports to its parent
983          * using a signal other than SIGCHLD, or a non-leader thread which
984          * we can only see if it is traced by us.
985          */
986         if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
987                 return 0;
988
989         return 1;
990 }
991
992 /*
993  * Handle sys_wait4 work for one task in state EXIT_ZOMBIE.  We hold
994  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
995  * the lock and this task is uninteresting.  If we return nonzero, we have
996  * released the lock and the system call should return.
997  */
998 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
999 {
1000         int state, status;
1001         pid_t pid = task_pid_vnr(p);
1002         uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1003         struct waitid_info *infop;
1004
1005         if (!likely(wo->wo_flags & WEXITED))
1006                 return 0;
1007
1008         if (unlikely(wo->wo_flags & WNOWAIT)) {
1009                 status = p->exit_code;
1010                 get_task_struct(p);
1011                 read_unlock(&tasklist_lock);
1012                 sched_annotate_sleep();
1013                 if (wo->wo_rusage)
1014                         getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1015                 put_task_struct(p);
1016                 goto out_info;
1017         }
1018         /*
1019          * Move the task's state to DEAD/TRACE, only one thread can do this.
1020          */
1021         state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1022                 EXIT_TRACE : EXIT_DEAD;
1023         if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1024                 return 0;
1025         /*
1026          * We own this thread, nobody else can reap it.
1027          */
1028         read_unlock(&tasklist_lock);
1029         sched_annotate_sleep();
1030
1031         /*
1032          * Check thread_group_leader() to exclude the traced sub-threads.
1033          */
1034         if (state == EXIT_DEAD && thread_group_leader(p)) {
1035                 struct signal_struct *sig = p->signal;
1036                 struct signal_struct *psig = current->signal;
1037                 unsigned long maxrss;
1038                 u64 tgutime, tgstime;
1039
1040                 /*
1041                  * The resource counters for the group leader are in its
1042                  * own task_struct.  Those for dead threads in the group
1043                  * are in its signal_struct, as are those for the child
1044                  * processes it has previously reaped.  All these
1045                  * accumulate in the parent's signal_struct c* fields.
1046                  *
1047                  * We don't bother to take a lock here to protect these
1048                  * p->signal fields because the whole thread group is dead
1049                  * and nobody can change them.
1050                  *
1051                  * psig->stats_lock also protects us from our sub-theads
1052                  * which can reap other children at the same time. Until
1053                  * we change k_getrusage()-like users to rely on this lock
1054                  * we have to take ->siglock as well.
1055                  *
1056                  * We use thread_group_cputime_adjusted() to get times for
1057                  * the thread group, which consolidates times for all threads
1058                  * in the group including the group leader.
1059                  */
1060                 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1061                 spin_lock_irq(&current->sighand->siglock);
1062                 write_seqlock(&psig->stats_lock);
1063                 psig->cutime += tgutime + sig->cutime;
1064                 psig->cstime += tgstime + sig->cstime;
1065                 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1066                 psig->cmin_flt +=
1067                         p->min_flt + sig->min_flt + sig->cmin_flt;
1068                 psig->cmaj_flt +=
1069                         p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1070                 psig->cnvcsw +=
1071                         p->nvcsw + sig->nvcsw + sig->cnvcsw;
1072                 psig->cnivcsw +=
1073                         p->nivcsw + sig->nivcsw + sig->cnivcsw;
1074                 psig->cinblock +=
1075                         task_io_get_inblock(p) +
1076                         sig->inblock + sig->cinblock;
1077                 psig->coublock +=
1078                         task_io_get_oublock(p) +
1079                         sig->oublock + sig->coublock;
1080                 maxrss = max(sig->maxrss, sig->cmaxrss);
1081                 if (psig->cmaxrss < maxrss)
1082                         psig->cmaxrss = maxrss;
1083                 task_io_accounting_add(&psig->ioac, &p->ioac);
1084                 task_io_accounting_add(&psig->ioac, &sig->ioac);
1085                 write_sequnlock(&psig->stats_lock);
1086                 spin_unlock_irq(&current->sighand->siglock);
1087         }
1088
1089         if (wo->wo_rusage)
1090                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1091         status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1092                 ? p->signal->group_exit_code : p->exit_code;
1093         wo->wo_stat = status;
1094
1095         if (state == EXIT_TRACE) {
1096                 write_lock_irq(&tasklist_lock);
1097                 /* We dropped tasklist, ptracer could die and untrace */
1098                 ptrace_unlink(p);
1099
1100                 /* If parent wants a zombie, don't release it now */
1101                 state = EXIT_ZOMBIE;
1102                 if (do_notify_parent(p, p->exit_signal))
1103                         state = EXIT_DEAD;
1104                 p->exit_state = state;
1105                 write_unlock_irq(&tasklist_lock);
1106         }
1107         if (state == EXIT_DEAD)
1108                 release_task(p);
1109
1110 out_info:
1111         infop = wo->wo_info;
1112         if (infop) {
1113                 if ((status & 0x7f) == 0) {
1114                         infop->cause = CLD_EXITED;
1115                         infop->status = status >> 8;
1116                 } else {
1117                         infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1118                         infop->status = status & 0x7f;
1119                 }
1120                 infop->pid = pid;
1121                 infop->uid = uid;
1122         }
1123
1124         return pid;
1125 }
1126
1127 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1128 {
1129         if (ptrace) {
1130                 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1131                         return &p->exit_code;
1132         } else {
1133                 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1134                         return &p->signal->group_exit_code;
1135         }
1136         return NULL;
1137 }
1138
1139 /**
1140  * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1141  * @wo: wait options
1142  * @ptrace: is the wait for ptrace
1143  * @p: task to wait for
1144  *
1145  * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1146  *
1147  * CONTEXT:
1148  * read_lock(&tasklist_lock), which is released if return value is
1149  * non-zero.  Also, grabs and releases @p->sighand->siglock.
1150  *
1151  * RETURNS:
1152  * 0 if wait condition didn't exist and search for other wait conditions
1153  * should continue.  Non-zero return, -errno on failure and @p's pid on
1154  * success, implies that tasklist_lock is released and wait condition
1155  * search should terminate.
1156  */
1157 static int wait_task_stopped(struct wait_opts *wo,
1158                                 int ptrace, struct task_struct *p)
1159 {
1160         struct waitid_info *infop;
1161         int exit_code, *p_code, why;
1162         uid_t uid = 0; /* unneeded, required by compiler */
1163         pid_t pid;
1164
1165         /*
1166          * Traditionally we see ptrace'd stopped tasks regardless of options.
1167          */
1168         if (!ptrace && !(wo->wo_flags & WUNTRACED))
1169                 return 0;
1170
1171         if (!task_stopped_code(p, ptrace))
1172                 return 0;
1173
1174         exit_code = 0;
1175         spin_lock_irq(&p->sighand->siglock);
1176
1177         p_code = task_stopped_code(p, ptrace);
1178         if (unlikely(!p_code))
1179                 goto unlock_sig;
1180
1181         exit_code = *p_code;
1182         if (!exit_code)
1183                 goto unlock_sig;
1184
1185         if (!unlikely(wo->wo_flags & WNOWAIT))
1186                 *p_code = 0;
1187
1188         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1189 unlock_sig:
1190         spin_unlock_irq(&p->sighand->siglock);
1191         if (!exit_code)
1192                 return 0;
1193
1194         /*
1195          * Now we are pretty sure this task is interesting.
1196          * Make sure it doesn't get reaped out from under us while we
1197          * give up the lock and then examine it below.  We don't want to
1198          * keep holding onto the tasklist_lock while we call getrusage and
1199          * possibly take page faults for user memory.
1200          */
1201         get_task_struct(p);
1202         pid = task_pid_vnr(p);
1203         why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1204         read_unlock(&tasklist_lock);
1205         sched_annotate_sleep();
1206         if (wo->wo_rusage)
1207                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1208         put_task_struct(p);
1209
1210         if (likely(!(wo->wo_flags & WNOWAIT)))
1211                 wo->wo_stat = (exit_code << 8) | 0x7f;
1212
1213         infop = wo->wo_info;
1214         if (infop) {
1215                 infop->cause = why;
1216                 infop->status = exit_code;
1217                 infop->pid = pid;
1218                 infop->uid = uid;
1219         }
1220         return pid;
1221 }
1222
1223 /*
1224  * Handle do_wait work for one task in a live, non-stopped state.
1225  * read_lock(&tasklist_lock) on entry.  If we return zero, we still hold
1226  * the lock and this task is uninteresting.  If we return nonzero, we have
1227  * released the lock and the system call should return.
1228  */
1229 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1230 {
1231         struct waitid_info *infop;
1232         pid_t pid;
1233         uid_t uid;
1234
1235         if (!unlikely(wo->wo_flags & WCONTINUED))
1236                 return 0;
1237
1238         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1239                 return 0;
1240
1241         spin_lock_irq(&p->sighand->siglock);
1242         /* Re-check with the lock held.  */
1243         if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1244                 spin_unlock_irq(&p->sighand->siglock);
1245                 return 0;
1246         }
1247         if (!unlikely(wo->wo_flags & WNOWAIT))
1248                 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1249         uid = from_kuid_munged(current_user_ns(), task_uid(p));
1250         spin_unlock_irq(&p->sighand->siglock);
1251
1252         pid = task_pid_vnr(p);
1253         get_task_struct(p);
1254         read_unlock(&tasklist_lock);
1255         sched_annotate_sleep();
1256         if (wo->wo_rusage)
1257                 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1258         put_task_struct(p);
1259
1260         infop = wo->wo_info;
1261         if (!infop) {
1262                 wo->wo_stat = 0xffff;
1263         } else {
1264                 infop->cause = CLD_CONTINUED;
1265                 infop->pid = pid;
1266                 infop->uid = uid;
1267                 infop->status = SIGCONT;
1268         }
1269         return pid;
1270 }
1271
1272 /*
1273  * Consider @p for a wait by @parent.
1274  *
1275  * -ECHILD should be in ->notask_error before the first call.
1276  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1277  * Returns zero if the search for a child should continue;
1278  * then ->notask_error is 0 if @p is an eligible child,
1279  * or still -ECHILD.
1280  */
1281 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1282                                 struct task_struct *p)
1283 {
1284         /*
1285          * We can race with wait_task_zombie() from another thread.
1286          * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1287          * can't confuse the checks below.
1288          */
1289         int exit_state = READ_ONCE(p->exit_state);
1290         int ret;
1291
1292         if (unlikely(exit_state == EXIT_DEAD))
1293                 return 0;
1294
1295         ret = eligible_child(wo, ptrace, p);
1296         if (!ret)
1297                 return ret;
1298
1299         if (unlikely(exit_state == EXIT_TRACE)) {
1300                 /*
1301                  * ptrace == 0 means we are the natural parent. In this case
1302                  * we should clear notask_error, debugger will notify us.
1303                  */
1304                 if (likely(!ptrace))
1305                         wo->notask_error = 0;
1306                 return 0;
1307         }
1308
1309         if (likely(!ptrace) && unlikely(p->ptrace)) {
1310                 /*
1311                  * If it is traced by its real parent's group, just pretend
1312                  * the caller is ptrace_do_wait() and reap this child if it
1313                  * is zombie.
1314                  *
1315                  * This also hides group stop state from real parent; otherwise
1316                  * a single stop can be reported twice as group and ptrace stop.
1317                  * If a ptracer wants to distinguish these two events for its
1318                  * own children it should create a separate process which takes
1319                  * the role of real parent.
1320                  */
1321                 if (!ptrace_reparented(p))
1322                         ptrace = 1;
1323         }
1324
1325         /* slay zombie? */
1326         if (exit_state == EXIT_ZOMBIE) {
1327                 /* we don't reap group leaders with subthreads */
1328                 if (!delay_group_leader(p)) {
1329                         /*
1330                          * A zombie ptracee is only visible to its ptracer.
1331                          * Notification and reaping will be cascaded to the
1332                          * real parent when the ptracer detaches.
1333                          */
1334                         if (unlikely(ptrace) || likely(!p->ptrace))
1335                                 return wait_task_zombie(wo, p);
1336                 }
1337
1338                 /*
1339                  * Allow access to stopped/continued state via zombie by
1340                  * falling through.  Clearing of notask_error is complex.
1341                  *
1342                  * When !@ptrace:
1343                  *
1344                  * If WEXITED is set, notask_error should naturally be
1345                  * cleared.  If not, subset of WSTOPPED|WCONTINUED is set,
1346                  * so, if there are live subthreads, there are events to
1347                  * wait for.  If all subthreads are dead, it's still safe
1348                  * to clear - this function will be called again in finite
1349                  * amount time once all the subthreads are released and
1350                  * will then return without clearing.
1351                  *
1352                  * When @ptrace:
1353                  *
1354                  * Stopped state is per-task and thus can't change once the
1355                  * target task dies.  Only continued and exited can happen.
1356                  * Clear notask_error if WCONTINUED | WEXITED.
1357                  */
1358                 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1359                         wo->notask_error = 0;
1360         } else {
1361                 /*
1362                  * @p is alive and it's gonna stop, continue or exit, so
1363                  * there always is something to wait for.
1364                  */
1365                 wo->notask_error = 0;
1366         }
1367
1368         /*
1369          * Wait for stopped.  Depending on @ptrace, different stopped state
1370          * is used and the two don't interact with each other.
1371          */
1372         ret = wait_task_stopped(wo, ptrace, p);
1373         if (ret)
1374                 return ret;
1375
1376         /*
1377          * Wait for continued.  There's only one continued state and the
1378          * ptracer can consume it which can confuse the real parent.  Don't
1379          * use WCONTINUED from ptracer.  You don't need or want it.
1380          */
1381         return wait_task_continued(wo, p);
1382 }
1383
1384 /*
1385  * Do the work of do_wait() for one thread in the group, @tsk.
1386  *
1387  * -ECHILD should be in ->notask_error before the first call.
1388  * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1389  * Returns zero if the search for a child should continue; then
1390  * ->notask_error is 0 if there were any eligible children,
1391  * or still -ECHILD.
1392  */
1393 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1394 {
1395         struct task_struct *p;
1396
1397         list_for_each_entry(p, &tsk->children, sibling) {
1398                 int ret = wait_consider_task(wo, 0, p);
1399
1400                 if (ret)
1401                         return ret;
1402         }
1403
1404         return 0;
1405 }
1406
1407 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1408 {
1409         struct task_struct *p;
1410
1411         list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1412                 int ret = wait_consider_task(wo, 1, p);
1413
1414                 if (ret)
1415                         return ret;
1416         }
1417
1418         return 0;
1419 }
1420
1421 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1422                                 int sync, void *key)
1423 {
1424         struct wait_opts *wo = container_of(wait, struct wait_opts,
1425                                                 child_wait);
1426         struct task_struct *p = key;
1427
1428         if (!eligible_pid(wo, p))
1429                 return 0;
1430
1431         if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1432                 return 0;
1433
1434         return default_wake_function(wait, mode, sync, key);
1435 }
1436
1437 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1438 {
1439         __wake_up_sync_key(&parent->signal->wait_chldexit,
1440                            TASK_INTERRUPTIBLE, p);
1441 }
1442
1443 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1444                                  struct task_struct *target)
1445 {
1446         struct task_struct *parent =
1447                 !ptrace ? target->real_parent : target->parent;
1448
1449         return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1450                                      same_thread_group(current, parent));
1451 }
1452
1453 /*
1454  * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1455  * and tracee lists to find the target task.
1456  */
1457 static int do_wait_pid(struct wait_opts *wo)
1458 {
1459         bool ptrace;
1460         struct task_struct *target;
1461         int retval;
1462
1463         ptrace = false;
1464         target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1465         if (target && is_effectively_child(wo, ptrace, target)) {
1466                 retval = wait_consider_task(wo, ptrace, target);
1467                 if (retval)
1468                         return retval;
1469         }
1470
1471         ptrace = true;
1472         target = pid_task(wo->wo_pid, PIDTYPE_PID);
1473         if (target && target->ptrace &&
1474             is_effectively_child(wo, ptrace, target)) {
1475                 retval = wait_consider_task(wo, ptrace, target);
1476                 if (retval)
1477                         return retval;
1478         }
1479
1480         return 0;
1481 }
1482
1483 static long do_wait(struct wait_opts *wo)
1484 {
1485         int retval;
1486
1487         trace_sched_process_wait(wo->wo_pid);
1488
1489         init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1490         wo->child_wait.private = current;
1491         add_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1492 repeat:
1493         /*
1494          * If there is nothing that can match our criteria, just get out.
1495          * We will clear ->notask_error to zero if we see any child that
1496          * might later match our criteria, even if we are not able to reap
1497          * it yet.
1498          */
1499         wo->notask_error = -ECHILD;
1500         if ((wo->wo_type < PIDTYPE_MAX) &&
1501            (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1502                 goto notask;
1503
1504         set_current_state(TASK_INTERRUPTIBLE);
1505         read_lock(&tasklist_lock);
1506
1507         if (wo->wo_type == PIDTYPE_PID) {
1508                 retval = do_wait_pid(wo);
1509                 if (retval)
1510                         goto end;
1511         } else {
1512                 struct task_struct *tsk = current;
1513
1514                 do {
1515                         retval = do_wait_thread(wo, tsk);
1516                         if (retval)
1517                                 goto end;
1518
1519                         retval = ptrace_do_wait(wo, tsk);
1520                         if (retval)
1521                                 goto end;
1522
1523                         if (wo->wo_flags & __WNOTHREAD)
1524                                 break;
1525                 } while_each_thread(current, tsk);
1526         }
1527         read_unlock(&tasklist_lock);
1528
1529 notask:
1530         retval = wo->notask_error;
1531         if (!retval && !(wo->wo_flags & WNOHANG)) {
1532                 retval = -ERESTARTSYS;
1533                 if (!signal_pending(current)) {
1534                         schedule();
1535                         goto repeat;
1536                 }
1537         }
1538 end:
1539         __set_current_state(TASK_RUNNING);
1540         remove_wait_queue(&current->signal->wait_chldexit, &wo->child_wait);
1541         return retval;
1542 }
1543
1544 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1545                           int options, struct rusage *ru)
1546 {
1547         struct wait_opts wo;
1548         struct pid *pid = NULL;
1549         enum pid_type type;
1550         long ret;
1551         unsigned int f_flags = 0;
1552
1553         if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1554                         __WNOTHREAD|__WCLONE|__WALL))
1555                 return -EINVAL;
1556         if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1557                 return -EINVAL;
1558
1559         switch (which) {
1560         case P_ALL:
1561                 type = PIDTYPE_MAX;
1562                 break;
1563         case P_PID:
1564                 type = PIDTYPE_PID;
1565                 if (upid <= 0)
1566                         return -EINVAL;
1567
1568                 pid = find_get_pid(upid);
1569                 break;
1570         case P_PGID:
1571                 type = PIDTYPE_PGID;
1572                 if (upid < 0)
1573                         return -EINVAL;
1574
1575                 if (upid)
1576                         pid = find_get_pid(upid);
1577                 else
1578                         pid = get_task_pid(current, PIDTYPE_PGID);
1579                 break;
1580         case P_PIDFD:
1581                 type = PIDTYPE_PID;
1582                 if (upid < 0)
1583                         return -EINVAL;
1584
1585                 pid = pidfd_get_pid(upid, &f_flags);
1586                 if (IS_ERR(pid))
1587                         return PTR_ERR(pid);
1588
1589                 break;
1590         default:
1591                 return -EINVAL;
1592         }
1593
1594         wo.wo_type      = type;
1595         wo.wo_pid       = pid;
1596         wo.wo_flags     = options;
1597         wo.wo_info      = infop;
1598         wo.wo_rusage    = ru;
1599         if (f_flags & O_NONBLOCK)
1600                 wo.wo_flags |= WNOHANG;
1601
1602         ret = do_wait(&wo);
1603         if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1604                 ret = -EAGAIN;
1605
1606         put_pid(pid);
1607         return ret;
1608 }
1609
1610 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1611                 infop, int, options, struct rusage __user *, ru)
1612 {
1613         struct rusage r;
1614         struct waitid_info info = {.status = 0};
1615         long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1616         int signo = 0;
1617
1618         if (err > 0) {
1619                 signo = SIGCHLD;
1620                 err = 0;
1621                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1622                         return -EFAULT;
1623         }
1624         if (!infop)
1625                 return err;
1626
1627         if (!user_write_access_begin(infop, sizeof(*infop)))
1628                 return -EFAULT;
1629
1630         unsafe_put_user(signo, &infop->si_signo, Efault);
1631         unsafe_put_user(0, &infop->si_errno, Efault);
1632         unsafe_put_user(info.cause, &infop->si_code, Efault);
1633         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1634         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1635         unsafe_put_user(info.status, &infop->si_status, Efault);
1636         user_write_access_end();
1637         return err;
1638 Efault:
1639         user_write_access_end();
1640         return -EFAULT;
1641 }
1642
1643 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1644                   struct rusage *ru)
1645 {
1646         struct wait_opts wo;
1647         struct pid *pid = NULL;
1648         enum pid_type type;
1649         long ret;
1650
1651         if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1652                         __WNOTHREAD|__WCLONE|__WALL))
1653                 return -EINVAL;
1654
1655         /* -INT_MIN is not defined */
1656         if (upid == INT_MIN)
1657                 return -ESRCH;
1658
1659         if (upid == -1)
1660                 type = PIDTYPE_MAX;
1661         else if (upid < 0) {
1662                 type = PIDTYPE_PGID;
1663                 pid = find_get_pid(-upid);
1664         } else if (upid == 0) {
1665                 type = PIDTYPE_PGID;
1666                 pid = get_task_pid(current, PIDTYPE_PGID);
1667         } else /* upid > 0 */ {
1668                 type = PIDTYPE_PID;
1669                 pid = find_get_pid(upid);
1670         }
1671
1672         wo.wo_type      = type;
1673         wo.wo_pid       = pid;
1674         wo.wo_flags     = options | WEXITED;
1675         wo.wo_info      = NULL;
1676         wo.wo_stat      = 0;
1677         wo.wo_rusage    = ru;
1678         ret = do_wait(&wo);
1679         put_pid(pid);
1680         if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1681                 ret = -EFAULT;
1682
1683         return ret;
1684 }
1685
1686 int kernel_wait(pid_t pid, int *stat)
1687 {
1688         struct wait_opts wo = {
1689                 .wo_type        = PIDTYPE_PID,
1690                 .wo_pid         = find_get_pid(pid),
1691                 .wo_flags       = WEXITED,
1692         };
1693         int ret;
1694
1695         ret = do_wait(&wo);
1696         if (ret > 0 && wo.wo_stat)
1697                 *stat = wo.wo_stat;
1698         put_pid(wo.wo_pid);
1699         return ret;
1700 }
1701
1702 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1703                 int, options, struct rusage __user *, ru)
1704 {
1705         struct rusage r;
1706         long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1707
1708         if (err > 0) {
1709                 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1710                         return -EFAULT;
1711         }
1712         return err;
1713 }
1714
1715 #ifdef __ARCH_WANT_SYS_WAITPID
1716
1717 /*
1718  * sys_waitpid() remains for compatibility. waitpid() should be
1719  * implemented by calling sys_wait4() from libc.a.
1720  */
1721 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1722 {
1723         return kernel_wait4(pid, stat_addr, options, NULL);
1724 }
1725
1726 #endif
1727
1728 #ifdef CONFIG_COMPAT
1729 COMPAT_SYSCALL_DEFINE4(wait4,
1730         compat_pid_t, pid,
1731         compat_uint_t __user *, stat_addr,
1732         int, options,
1733         struct compat_rusage __user *, ru)
1734 {
1735         struct rusage r;
1736         long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1737         if (err > 0) {
1738                 if (ru && put_compat_rusage(&r, ru))
1739                         return -EFAULT;
1740         }
1741         return err;
1742 }
1743
1744 COMPAT_SYSCALL_DEFINE5(waitid,
1745                 int, which, compat_pid_t, pid,
1746                 struct compat_siginfo __user *, infop, int, options,
1747                 struct compat_rusage __user *, uru)
1748 {
1749         struct rusage ru;
1750         struct waitid_info info = {.status = 0};
1751         long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1752         int signo = 0;
1753         if (err > 0) {
1754                 signo = SIGCHLD;
1755                 err = 0;
1756                 if (uru) {
1757                         /* kernel_waitid() overwrites everything in ru */
1758                         if (COMPAT_USE_64BIT_TIME)
1759                                 err = copy_to_user(uru, &ru, sizeof(ru));
1760                         else
1761                                 err = put_compat_rusage(&ru, uru);
1762                         if (err)
1763                                 return -EFAULT;
1764                 }
1765         }
1766
1767         if (!infop)
1768                 return err;
1769
1770         if (!user_write_access_begin(infop, sizeof(*infop)))
1771                 return -EFAULT;
1772
1773         unsafe_put_user(signo, &infop->si_signo, Efault);
1774         unsafe_put_user(0, &infop->si_errno, Efault);
1775         unsafe_put_user(info.cause, &infop->si_code, Efault);
1776         unsafe_put_user(info.pid, &infop->si_pid, Efault);
1777         unsafe_put_user(info.uid, &infop->si_uid, Efault);
1778         unsafe_put_user(info.status, &infop->si_status, Efault);
1779         user_write_access_end();
1780         return err;
1781 Efault:
1782         user_write_access_end();
1783         return -EFAULT;
1784 }
1785 #endif
1786
1787 /**
1788  * thread_group_exited - check that a thread group has exited
1789  * @pid: tgid of thread group to be checked.
1790  *
1791  * Test if the thread group represented by tgid has exited (all
1792  * threads are zombies, dead or completely gone).
1793  *
1794  * Return: true if the thread group has exited. false otherwise.
1795  */
1796 bool thread_group_exited(struct pid *pid)
1797 {
1798         struct task_struct *task;
1799         bool exited;
1800
1801         rcu_read_lock();
1802         task = pid_task(pid, PIDTYPE_PID);
1803         exited = !task ||
1804                 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1805         rcu_read_unlock();
1806
1807         return exited;
1808 }
1809 EXPORT_SYMBOL(thread_group_exited);
1810
1811 __weak void abort(void)
1812 {
1813         BUG();
1814
1815         /* if that doesn't kill us, halt */
1816         panic("Oops failed to kill thread");
1817 }
1818 EXPORT_SYMBOL(abort);