Loading...
1/*
2 * linux/kernel/exit.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/mm.h>
8#include <linux/slab.h>
9#include <linux/interrupt.h>
10#include <linux/module.h>
11#include <linux/capability.h>
12#include <linux/completion.h>
13#include <linux/personality.h>
14#include <linux/tty.h>
15#include <linux/iocontext.h>
16#include <linux/key.h>
17#include <linux/security.h>
18#include <linux/cpu.h>
19#include <linux/acct.h>
20#include <linux/tsacct_kern.h>
21#include <linux/file.h>
22#include <linux/fdtable.h>
23#include <linux/freezer.h>
24#include <linux/binfmts.h>
25#include <linux/nsproxy.h>
26#include <linux/pid_namespace.h>
27#include <linux/ptrace.h>
28#include <linux/profile.h>
29#include <linux/mount.h>
30#include <linux/proc_fs.h>
31#include <linux/kthread.h>
32#include <linux/mempolicy.h>
33#include <linux/taskstats_kern.h>
34#include <linux/delayacct.h>
35#include <linux/cgroup.h>
36#include <linux/syscalls.h>
37#include <linux/signal.h>
38#include <linux/posix-timers.h>
39#include <linux/cn_proc.h>
40#include <linux/mutex.h>
41#include <linux/futex.h>
42#include <linux/pipe_fs_i.h>
43#include <linux/audit.h> /* for audit_free() */
44#include <linux/resource.h>
45#include <linux/blkdev.h>
46#include <linux/task_io_accounting_ops.h>
47#include <linux/tracehook.h>
48#include <linux/fs_struct.h>
49#include <linux/init_task.h>
50#include <linux/perf_event.h>
51#include <trace/events/sched.h>
52#include <linux/hw_breakpoint.h>
53#include <linux/oom.h>
54#include <linux/writeback.h>
55#include <linux/shm.h>
56#include <linux/kcov.h>
57#include <linux/random.h>
58
59#include <linux/uaccess.h>
60#include <asm/unistd.h>
61#include <asm/pgtable.h>
62#include <asm/mmu_context.h>
63
64static void __unhash_process(struct task_struct *p, bool group_dead)
65{
66 nr_threads--;
67 detach_pid(p, PIDTYPE_PID);
68 if (group_dead) {
69 detach_pid(p, PIDTYPE_PGID);
70 detach_pid(p, PIDTYPE_SID);
71
72 list_del_rcu(&p->tasks);
73 list_del_init(&p->sibling);
74 __this_cpu_dec(process_counts);
75 }
76 list_del_rcu(&p->thread_group);
77 list_del_rcu(&p->thread_node);
78}
79
80/*
81 * This function expects the tasklist_lock write-locked.
82 */
83static void __exit_signal(struct task_struct *tsk)
84{
85 struct signal_struct *sig = tsk->signal;
86 bool group_dead = thread_group_leader(tsk);
87 struct sighand_struct *sighand;
88 struct tty_struct *uninitialized_var(tty);
89 cputime_t utime, stime;
90
91 sighand = rcu_dereference_check(tsk->sighand,
92 lockdep_tasklist_lock_is_held());
93 spin_lock(&sighand->siglock);
94
95#ifdef CONFIG_POSIX_TIMERS
96 posix_cpu_timers_exit(tsk);
97 if (group_dead) {
98 posix_cpu_timers_exit_group(tsk);
99 } else {
100 /*
101 * This can only happen if the caller is de_thread().
102 * FIXME: this is the temporary hack, we should teach
103 * posix-cpu-timers to handle this case correctly.
104 */
105 if (unlikely(has_group_leader_pid(tsk)))
106 posix_cpu_timers_exit_group(tsk);
107 }
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
167static 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
177void release_task(struct task_struct *p)
178{
179 struct task_struct *leader;
180 int zap_leader;
181repeat:
182 /* don't need to get the RCU readlock here - the process is dead and
183 * can't be modifying its own credentials. But shut RCU-lockdep up */
184 rcu_read_lock();
185 atomic_dec(&__task_cred(p)->user->processes);
186 rcu_read_unlock();
187
188 proc_flush_task(p);
189
190 write_lock_irq(&tasklist_lock);
191 ptrace_release_task(p);
192 __exit_signal(p);
193
194 /*
195 * If we are the last non-leader member of the thread
196 * group, and the leader is zombie, then notify the
197 * group leader's parent process. (if it wants notification.)
198 */
199 zap_leader = 0;
200 leader = p->group_leader;
201 if (leader != p && thread_group_empty(leader)
202 && leader->exit_state == EXIT_ZOMBIE) {
203 /*
204 * If we were the last child thread and the leader has
205 * exited already, and the leader's parent ignores SIGCHLD,
206 * then we are the one who should release the leader.
207 */
208 zap_leader = do_notify_parent(leader, leader->exit_signal);
209 if (zap_leader)
210 leader->exit_state = EXIT_DEAD;
211 }
212
213 write_unlock_irq(&tasklist_lock);
214 release_thread(p);
215 call_rcu(&p->rcu, delayed_put_task_struct);
216
217 p = leader;
218 if (unlikely(zap_leader))
219 goto repeat;
220}
221
222/*
223 * Note that if this function returns a valid task_struct pointer (!NULL)
224 * task->usage must remain >0 for the duration of the RCU critical section.
225 */
226struct task_struct *task_rcu_dereference(struct task_struct **ptask)
227{
228 struct sighand_struct *sighand;
229 struct task_struct *task;
230
231 /*
232 * We need to verify that release_task() was not called and thus
233 * delayed_put_task_struct() can't run and drop the last reference
234 * before rcu_read_unlock(). We check task->sighand != NULL,
235 * but we can read the already freed and reused memory.
236 */
237retry:
238 task = rcu_dereference(*ptask);
239 if (!task)
240 return NULL;
241
242 probe_kernel_address(&task->sighand, sighand);
243
244 /*
245 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
246 * was already freed we can not miss the preceding update of this
247 * pointer.
248 */
249 smp_rmb();
250 if (unlikely(task != READ_ONCE(*ptask)))
251 goto retry;
252
253 /*
254 * We've re-checked that "task == *ptask", now we have two different
255 * cases:
256 *
257 * 1. This is actually the same task/task_struct. In this case
258 * sighand != NULL tells us it is still alive.
259 *
260 * 2. This is another task which got the same memory for task_struct.
261 * We can't know this of course, and we can not trust
262 * sighand != NULL.
263 *
264 * In this case we actually return a random value, but this is
265 * correct.
266 *
267 * If we return NULL - we can pretend that we actually noticed that
268 * *ptask was updated when the previous task has exited. Or pretend
269 * that probe_slab_address(&sighand) reads NULL.
270 *
271 * If we return the new task (because sighand is not NULL for any
272 * reason) - this is fine too. This (new) task can't go away before
273 * another gp pass.
274 *
275 * And note: We could even eliminate the false positive if re-read
276 * task->sighand once again to avoid the falsely NULL. But this case
277 * is very unlikely so we don't care.
278 */
279 if (!sighand)
280 return NULL;
281
282 return task;
283}
284
285struct task_struct *try_get_task_struct(struct task_struct **ptask)
286{
287 struct task_struct *task;
288
289 rcu_read_lock();
290 task = task_rcu_dereference(ptask);
291 if (task)
292 get_task_struct(task);
293 rcu_read_unlock();
294
295 return task;
296}
297
298/*
299 * Determine if a process group is "orphaned", according to the POSIX
300 * definition in 2.2.2.52. Orphaned process groups are not to be affected
301 * by terminal-generated stop signals. Newly orphaned process groups are
302 * to receive a SIGHUP and a SIGCONT.
303 *
304 * "I ask you, have you ever known what it is to be an orphan?"
305 */
306static int will_become_orphaned_pgrp(struct pid *pgrp,
307 struct task_struct *ignored_task)
308{
309 struct task_struct *p;
310
311 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
312 if ((p == ignored_task) ||
313 (p->exit_state && thread_group_empty(p)) ||
314 is_global_init(p->real_parent))
315 continue;
316
317 if (task_pgrp(p->real_parent) != pgrp &&
318 task_session(p->real_parent) == task_session(p))
319 return 0;
320 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
321
322 return 1;
323}
324
325int is_current_pgrp_orphaned(void)
326{
327 int retval;
328
329 read_lock(&tasklist_lock);
330 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
331 read_unlock(&tasklist_lock);
332
333 return retval;
334}
335
336static bool has_stopped_jobs(struct pid *pgrp)
337{
338 struct task_struct *p;
339
340 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
341 if (p->signal->flags & SIGNAL_STOP_STOPPED)
342 return true;
343 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
344
345 return false;
346}
347
348/*
349 * Check to see if any process groups have become orphaned as
350 * a result of our exiting, and if they have any stopped jobs,
351 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
352 */
353static void
354kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
355{
356 struct pid *pgrp = task_pgrp(tsk);
357 struct task_struct *ignored_task = tsk;
358
359 if (!parent)
360 /* exit: our father is in a different pgrp than
361 * we are and we were the only connection outside.
362 */
363 parent = tsk->real_parent;
364 else
365 /* reparent: our child is in a different pgrp than
366 * we are, and it was the only connection outside.
367 */
368 ignored_task = NULL;
369
370 if (task_pgrp(parent) != pgrp &&
371 task_session(parent) == task_session(tsk) &&
372 will_become_orphaned_pgrp(pgrp, ignored_task) &&
373 has_stopped_jobs(pgrp)) {
374 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
375 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
376 }
377}
378
379#ifdef CONFIG_MEMCG
380/*
381 * A task is exiting. If it owned this mm, find a new owner for the mm.
382 */
383void mm_update_next_owner(struct mm_struct *mm)
384{
385 struct task_struct *c, *g, *p = current;
386
387retry:
388 /*
389 * If the exiting or execing task is not the owner, it's
390 * someone else's problem.
391 */
392 if (mm->owner != p)
393 return;
394 /*
395 * The current owner is exiting/execing and there are no other
396 * candidates. Do not leave the mm pointing to a possibly
397 * freed task structure.
398 */
399 if (atomic_read(&mm->mm_users) <= 1) {
400 mm->owner = NULL;
401 return;
402 }
403
404 read_lock(&tasklist_lock);
405 /*
406 * Search in the children
407 */
408 list_for_each_entry(c, &p->children, sibling) {
409 if (c->mm == mm)
410 goto assign_new_owner;
411 }
412
413 /*
414 * Search in the siblings
415 */
416 list_for_each_entry(c, &p->real_parent->children, sibling) {
417 if (c->mm == mm)
418 goto assign_new_owner;
419 }
420
421 /*
422 * Search through everything else, we should not get here often.
423 */
424 for_each_process(g) {
425 if (g->flags & PF_KTHREAD)
426 continue;
427 for_each_thread(g, c) {
428 if (c->mm == mm)
429 goto assign_new_owner;
430 if (c->mm)
431 break;
432 }
433 }
434 read_unlock(&tasklist_lock);
435 /*
436 * We found no owner yet mm_users > 1: this implies that we are
437 * most likely racing with swapoff (try_to_unuse()) or /proc or
438 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
439 */
440 mm->owner = NULL;
441 return;
442
443assign_new_owner:
444 BUG_ON(c == p);
445 get_task_struct(c);
446 /*
447 * The task_lock protects c->mm from changing.
448 * We always want mm->owner->mm == mm
449 */
450 task_lock(c);
451 /*
452 * Delay read_unlock() till we have the task_lock()
453 * to ensure that c does not slip away underneath us
454 */
455 read_unlock(&tasklist_lock);
456 if (c->mm != mm) {
457 task_unlock(c);
458 put_task_struct(c);
459 goto retry;
460 }
461 mm->owner = c;
462 task_unlock(c);
463 put_task_struct(c);
464}
465#endif /* CONFIG_MEMCG */
466
467/*
468 * Turn us into a lazy TLB process if we
469 * aren't already..
470 */
471static void exit_mm(struct task_struct *tsk)
472{
473 struct mm_struct *mm = tsk->mm;
474 struct core_state *core_state;
475
476 mm_release(tsk, mm);
477 if (!mm)
478 return;
479 sync_mm_rss(mm);
480 /*
481 * Serialize with any possible pending coredump.
482 * We must hold mmap_sem around checking core_state
483 * and clearing tsk->mm. The core-inducing thread
484 * will increment ->nr_threads for each thread in the
485 * group with ->mm != NULL.
486 */
487 down_read(&mm->mmap_sem);
488 core_state = mm->core_state;
489 if (core_state) {
490 struct core_thread self;
491
492 up_read(&mm->mmap_sem);
493
494 self.task = tsk;
495 self.next = xchg(&core_state->dumper.next, &self);
496 /*
497 * Implies mb(), the result of xchg() must be visible
498 * to core_state->dumper.
499 */
500 if (atomic_dec_and_test(&core_state->nr_threads))
501 complete(&core_state->startup);
502
503 for (;;) {
504 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
505 if (!self.task) /* see coredump_finish() */
506 break;
507 freezable_schedule();
508 }
509 __set_task_state(tsk, TASK_RUNNING);
510 down_read(&mm->mmap_sem);
511 }
512 atomic_inc(&mm->mm_count);
513 BUG_ON(mm != tsk->active_mm);
514 /* more a memory barrier than a real lock */
515 task_lock(tsk);
516 tsk->mm = NULL;
517 up_read(&mm->mmap_sem);
518 enter_lazy_tlb(mm, current);
519 task_unlock(tsk);
520 mm_update_next_owner(mm);
521 mmput(mm);
522 if (test_thread_flag(TIF_MEMDIE))
523 exit_oom_victim();
524}
525
526static struct task_struct *find_alive_thread(struct task_struct *p)
527{
528 struct task_struct *t;
529
530 for_each_thread(p, t) {
531 if (!(t->flags & PF_EXITING))
532 return t;
533 }
534 return NULL;
535}
536
537static struct task_struct *find_child_reaper(struct task_struct *father)
538 __releases(&tasklist_lock)
539 __acquires(&tasklist_lock)
540{
541 struct pid_namespace *pid_ns = task_active_pid_ns(father);
542 struct task_struct *reaper = pid_ns->child_reaper;
543
544 if (likely(reaper != father))
545 return reaper;
546
547 reaper = find_alive_thread(father);
548 if (reaper) {
549 pid_ns->child_reaper = reaper;
550 return reaper;
551 }
552
553 write_unlock_irq(&tasklist_lock);
554 if (unlikely(pid_ns == &init_pid_ns)) {
555 panic("Attempted to kill init! exitcode=0x%08x\n",
556 father->signal->group_exit_code ?: father->exit_code);
557 }
558 zap_pid_ns_processes(pid_ns);
559 write_lock_irq(&tasklist_lock);
560
561 return father;
562}
563
564/*
565 * When we die, we re-parent all our children, and try to:
566 * 1. give them to another thread in our thread group, if such a member exists
567 * 2. give it to the first ancestor process which prctl'd itself as a
568 * child_subreaper for its children (like a service manager)
569 * 3. give it to the init process (PID 1) in our pid namespace
570 */
571static struct task_struct *find_new_reaper(struct task_struct *father,
572 struct task_struct *child_reaper)
573{
574 struct task_struct *thread, *reaper;
575
576 thread = find_alive_thread(father);
577 if (thread)
578 return thread;
579
580 if (father->signal->has_child_subreaper) {
581 /*
582 * Find the first ->is_child_subreaper ancestor in our pid_ns.
583 * We start from father to ensure we can not look into another
584 * namespace, this is safe because all its threads are dead.
585 */
586 for (reaper = father;
587 !same_thread_group(reaper, child_reaper);
588 reaper = reaper->real_parent) {
589 /* call_usermodehelper() descendants need this check */
590 if (reaper == &init_task)
591 break;
592 if (!reaper->signal->is_child_subreaper)
593 continue;
594 thread = find_alive_thread(reaper);
595 if (thread)
596 return thread;
597 }
598 }
599
600 return child_reaper;
601}
602
603/*
604* Any that need to be release_task'd are put on the @dead list.
605 */
606static void reparent_leader(struct task_struct *father, struct task_struct *p,
607 struct list_head *dead)
608{
609 if (unlikely(p->exit_state == EXIT_DEAD))
610 return;
611
612 /* We don't want people slaying init. */
613 p->exit_signal = SIGCHLD;
614
615 /* If it has exited notify the new parent about this child's death. */
616 if (!p->ptrace &&
617 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
618 if (do_notify_parent(p, p->exit_signal)) {
619 p->exit_state = EXIT_DEAD;
620 list_add(&p->ptrace_entry, dead);
621 }
622 }
623
624 kill_orphaned_pgrp(p, father);
625}
626
627/*
628 * This does two things:
629 *
630 * A. Make init inherit all the child processes
631 * B. Check to see if any process groups have become orphaned
632 * as a result of our exiting, and if they have any stopped
633 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
634 */
635static void forget_original_parent(struct task_struct *father,
636 struct list_head *dead)
637{
638 struct task_struct *p, *t, *reaper;
639
640 if (unlikely(!list_empty(&father->ptraced)))
641 exit_ptrace(father, dead);
642
643 /* Can drop and reacquire tasklist_lock */
644 reaper = find_child_reaper(father);
645 if (list_empty(&father->children))
646 return;
647
648 reaper = find_new_reaper(father, reaper);
649 list_for_each_entry(p, &father->children, sibling) {
650 for_each_thread(p, t) {
651 t->real_parent = reaper;
652 BUG_ON((!t->ptrace) != (t->parent == father));
653 if (likely(!t->ptrace))
654 t->parent = t->real_parent;
655 if (t->pdeath_signal)
656 group_send_sig_info(t->pdeath_signal,
657 SEND_SIG_NOINFO, t);
658 }
659 /*
660 * If this is a threaded reparent there is no need to
661 * notify anyone anything has happened.
662 */
663 if (!same_thread_group(reaper, father))
664 reparent_leader(father, p, dead);
665 }
666 list_splice_tail_init(&father->children, &reaper->children);
667}
668
669/*
670 * Send signals to all our closest relatives so that they know
671 * to properly mourn us..
672 */
673static void exit_notify(struct task_struct *tsk, int group_dead)
674{
675 bool autoreap;
676 struct task_struct *p, *n;
677 LIST_HEAD(dead);
678
679 write_lock_irq(&tasklist_lock);
680 forget_original_parent(tsk, &dead);
681
682 if (group_dead)
683 kill_orphaned_pgrp(tsk->group_leader, NULL);
684
685 if (unlikely(tsk->ptrace)) {
686 int sig = thread_group_leader(tsk) &&
687 thread_group_empty(tsk) &&
688 !ptrace_reparented(tsk) ?
689 tsk->exit_signal : SIGCHLD;
690 autoreap = do_notify_parent(tsk, sig);
691 } else if (thread_group_leader(tsk)) {
692 autoreap = thread_group_empty(tsk) &&
693 do_notify_parent(tsk, tsk->exit_signal);
694 } else {
695 autoreap = true;
696 }
697
698 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
699 if (tsk->exit_state == EXIT_DEAD)
700 list_add(&tsk->ptrace_entry, &dead);
701
702 /* mt-exec, de_thread() is waiting for group leader */
703 if (unlikely(tsk->signal->notify_count < 0))
704 wake_up_process(tsk->signal->group_exit_task);
705 write_unlock_irq(&tasklist_lock);
706
707 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
708 list_del_init(&p->ptrace_entry);
709 release_task(p);
710 }
711}
712
713#ifdef CONFIG_DEBUG_STACK_USAGE
714static void check_stack_usage(void)
715{
716 static DEFINE_SPINLOCK(low_water_lock);
717 static int lowest_to_date = THREAD_SIZE;
718 unsigned long free;
719
720 free = stack_not_used(current);
721
722 if (free >= lowest_to_date)
723 return;
724
725 spin_lock(&low_water_lock);
726 if (free < lowest_to_date) {
727 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
728 current->comm, task_pid_nr(current), free);
729 lowest_to_date = free;
730 }
731 spin_unlock(&low_water_lock);
732}
733#else
734static inline void check_stack_usage(void) {}
735#endif
736
737void __noreturn do_exit(long code)
738{
739 struct task_struct *tsk = current;
740 int group_dead;
741 TASKS_RCU(int tasks_rcu_i);
742
743 profile_task_exit(tsk);
744 kcov_task_exit(tsk);
745
746 WARN_ON(blk_needs_flush_plug(tsk));
747
748 if (unlikely(in_interrupt()))
749 panic("Aiee, killing interrupt handler!");
750 if (unlikely(!tsk->pid))
751 panic("Attempted to kill the idle task!");
752
753 /*
754 * If do_exit is called because this processes oopsed, it's possible
755 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
756 * continuing. Amongst other possible reasons, this is to prevent
757 * mm_release()->clear_child_tid() from writing to a user-controlled
758 * kernel address.
759 */
760 set_fs(USER_DS);
761
762 ptrace_event(PTRACE_EVENT_EXIT, code);
763
764 validate_creds_for_do_exit(tsk);
765
766 /*
767 * We're taking recursive faults here in do_exit. Safest is to just
768 * leave this task alone and wait for reboot.
769 */
770 if (unlikely(tsk->flags & PF_EXITING)) {
771 pr_alert("Fixing recursive fault but reboot is needed!\n");
772 /*
773 * We can do this unlocked here. The futex code uses
774 * this flag just to verify whether the pi state
775 * cleanup has been done or not. In the worst case it
776 * loops once more. We pretend that the cleanup was
777 * done as there is no way to return. Either the
778 * OWNER_DIED bit is set by now or we push the blocked
779 * task into the wait for ever nirwana as well.
780 */
781 tsk->flags |= PF_EXITPIDONE;
782 set_current_state(TASK_UNINTERRUPTIBLE);
783 schedule();
784 }
785
786 exit_signals(tsk); /* sets PF_EXITING */
787 /*
788 * Ensure that all new tsk->pi_lock acquisitions must observe
789 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
790 */
791 smp_mb();
792 /*
793 * Ensure that we must observe the pi_state in exit_mm() ->
794 * mm_release() -> exit_pi_state_list().
795 */
796 raw_spin_unlock_wait(&tsk->pi_lock);
797
798 if (unlikely(in_atomic())) {
799 pr_info("note: %s[%d] exited with preempt_count %d\n",
800 current->comm, task_pid_nr(current),
801 preempt_count());
802 preempt_count_set(PREEMPT_ENABLED);
803 }
804
805 /* sync mm's RSS info before statistics gathering */
806 if (tsk->mm)
807 sync_mm_rss(tsk->mm);
808 acct_update_integrals(tsk);
809 group_dead = atomic_dec_and_test(&tsk->signal->live);
810 if (group_dead) {
811#ifdef CONFIG_POSIX_TIMERS
812 hrtimer_cancel(&tsk->signal->real_timer);
813 exit_itimers(tsk->signal);
814#endif
815 if (tsk->mm)
816 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
817 }
818 acct_collect(code, group_dead);
819 if (group_dead)
820 tty_audit_exit();
821 audit_free(tsk);
822
823 tsk->exit_code = code;
824 taskstats_exit(tsk, group_dead);
825
826 exit_mm(tsk);
827
828 if (group_dead)
829 acct_process();
830 trace_sched_process_exit(tsk);
831
832 exit_sem(tsk);
833 exit_shm(tsk);
834 exit_files(tsk);
835 exit_fs(tsk);
836 if (group_dead)
837 disassociate_ctty(1);
838 exit_task_namespaces(tsk);
839 exit_task_work(tsk);
840 exit_thread(tsk);
841
842 /*
843 * Flush inherited counters to the parent - before the parent
844 * gets woken up by child-exit notifications.
845 *
846 * because of cgroup mode, must be called before cgroup_exit()
847 */
848 perf_event_exit_task(tsk);
849
850 sched_autogroup_exit_task(tsk);
851 cgroup_exit(tsk);
852
853 /*
854 * FIXME: do that only when needed, using sched_exit tracepoint
855 */
856 flush_ptrace_hw_breakpoint(tsk);
857
858 TASKS_RCU(preempt_disable());
859 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
860 TASKS_RCU(preempt_enable());
861 exit_notify(tsk, group_dead);
862 proc_exit_connector(tsk);
863 mpol_put_task_policy(tsk);
864#ifdef CONFIG_FUTEX
865 if (unlikely(current->pi_state_cache))
866 kfree(current->pi_state_cache);
867#endif
868 /*
869 * Make sure we are holding no locks:
870 */
871 debug_check_no_locks_held();
872 /*
873 * We can do this unlocked here. The futex code uses this flag
874 * just to verify whether the pi state cleanup has been done
875 * or not. In the worst case it loops once more.
876 */
877 tsk->flags |= PF_EXITPIDONE;
878
879 if (tsk->io_context)
880 exit_io_context(tsk);
881
882 if (tsk->splice_pipe)
883 free_pipe_info(tsk->splice_pipe);
884
885 if (tsk->task_frag.page)
886 put_page(tsk->task_frag.page);
887
888 validate_creds_for_do_exit(tsk);
889
890 check_stack_usage();
891 preempt_disable();
892 if (tsk->nr_dirtied)
893 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
894 exit_rcu();
895 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
896
897 do_task_dead();
898}
899EXPORT_SYMBOL_GPL(do_exit);
900
901void complete_and_exit(struct completion *comp, long code)
902{
903 if (comp)
904 complete(comp);
905
906 do_exit(code);
907}
908EXPORT_SYMBOL(complete_and_exit);
909
910SYSCALL_DEFINE1(exit, int, error_code)
911{
912 do_exit((error_code&0xff)<<8);
913}
914
915/*
916 * Take down every thread in the group. This is called by fatal signals
917 * as well as by sys_exit_group (below).
918 */
919void
920do_group_exit(int exit_code)
921{
922 struct signal_struct *sig = current->signal;
923
924 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
925
926 if (signal_group_exit(sig))
927 exit_code = sig->group_exit_code;
928 else if (!thread_group_empty(current)) {
929 struct sighand_struct *const sighand = current->sighand;
930
931 spin_lock_irq(&sighand->siglock);
932 if (signal_group_exit(sig))
933 /* Another thread got here before we took the lock. */
934 exit_code = sig->group_exit_code;
935 else {
936 sig->group_exit_code = exit_code;
937 sig->flags = SIGNAL_GROUP_EXIT;
938 zap_other_threads(current);
939 }
940 spin_unlock_irq(&sighand->siglock);
941 }
942
943 do_exit(exit_code);
944 /* NOTREACHED */
945}
946
947/*
948 * this kills every thread in the thread group. Note that any externally
949 * wait4()-ing process will get the correct exit code - even if this
950 * thread is not the thread group leader.
951 */
952SYSCALL_DEFINE1(exit_group, int, error_code)
953{
954 do_group_exit((error_code & 0xff) << 8);
955 /* NOTREACHED */
956 return 0;
957}
958
959struct wait_opts {
960 enum pid_type wo_type;
961 int wo_flags;
962 struct pid *wo_pid;
963
964 struct siginfo __user *wo_info;
965 int __user *wo_stat;
966 struct rusage __user *wo_rusage;
967
968 wait_queue_t child_wait;
969 int notask_error;
970};
971
972static inline
973struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
974{
975 if (type != PIDTYPE_PID)
976 task = task->group_leader;
977 return task->pids[type].pid;
978}
979
980static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
981{
982 return wo->wo_type == PIDTYPE_MAX ||
983 task_pid_type(p, wo->wo_type) == wo->wo_pid;
984}
985
986static int
987eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
988{
989 if (!eligible_pid(wo, p))
990 return 0;
991
992 /*
993 * Wait for all children (clone and not) if __WALL is set or
994 * if it is traced by us.
995 */
996 if (ptrace || (wo->wo_flags & __WALL))
997 return 1;
998
999 /*
1000 * Otherwise, wait for clone children *only* if __WCLONE is set;
1001 * otherwise, wait for non-clone children *only*.
1002 *
1003 * Note: a "clone" child here is one that reports to its parent
1004 * using a signal other than SIGCHLD, or a non-leader thread which
1005 * we can only see if it is traced by us.
1006 */
1007 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1008 return 0;
1009
1010 return 1;
1011}
1012
1013static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1014 pid_t pid, uid_t uid, int why, int status)
1015{
1016 struct siginfo __user *infop;
1017 int retval = wo->wo_rusage
1018 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1019
1020 put_task_struct(p);
1021 infop = wo->wo_info;
1022 if (infop) {
1023 if (!retval)
1024 retval = put_user(SIGCHLD, &infop->si_signo);
1025 if (!retval)
1026 retval = put_user(0, &infop->si_errno);
1027 if (!retval)
1028 retval = put_user((short)why, &infop->si_code);
1029 if (!retval)
1030 retval = put_user(pid, &infop->si_pid);
1031 if (!retval)
1032 retval = put_user(uid, &infop->si_uid);
1033 if (!retval)
1034 retval = put_user(status, &infop->si_status);
1035 }
1036 if (!retval)
1037 retval = pid;
1038 return retval;
1039}
1040
1041/*
1042 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1043 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1044 * the lock and this task is uninteresting. If we return nonzero, we have
1045 * released the lock and the system call should return.
1046 */
1047static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1048{
1049 int state, retval, status;
1050 pid_t pid = task_pid_vnr(p);
1051 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1052 struct siginfo __user *infop;
1053
1054 if (!likely(wo->wo_flags & WEXITED))
1055 return 0;
1056
1057 if (unlikely(wo->wo_flags & WNOWAIT)) {
1058 int exit_code = p->exit_code;
1059 int why;
1060
1061 get_task_struct(p);
1062 read_unlock(&tasklist_lock);
1063 sched_annotate_sleep();
1064
1065 if ((exit_code & 0x7f) == 0) {
1066 why = CLD_EXITED;
1067 status = exit_code >> 8;
1068 } else {
1069 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1070 status = exit_code & 0x7f;
1071 }
1072 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1073 }
1074 /*
1075 * Move the task's state to DEAD/TRACE, only one thread can do this.
1076 */
1077 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1078 EXIT_TRACE : EXIT_DEAD;
1079 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1080 return 0;
1081 /*
1082 * We own this thread, nobody else can reap it.
1083 */
1084 read_unlock(&tasklist_lock);
1085 sched_annotate_sleep();
1086
1087 /*
1088 * Check thread_group_leader() to exclude the traced sub-threads.
1089 */
1090 if (state == EXIT_DEAD && thread_group_leader(p)) {
1091 struct signal_struct *sig = p->signal;
1092 struct signal_struct *psig = current->signal;
1093 unsigned long maxrss;
1094 cputime_t tgutime, tgstime;
1095
1096 /*
1097 * The resource counters for the group leader are in its
1098 * own task_struct. Those for dead threads in the group
1099 * are in its signal_struct, as are those for the child
1100 * processes it has previously reaped. All these
1101 * accumulate in the parent's signal_struct c* fields.
1102 *
1103 * We don't bother to take a lock here to protect these
1104 * p->signal fields because the whole thread group is dead
1105 * and nobody can change them.
1106 *
1107 * psig->stats_lock also protects us from our sub-theads
1108 * which can reap other children at the same time. Until
1109 * we change k_getrusage()-like users to rely on this lock
1110 * we have to take ->siglock as well.
1111 *
1112 * We use thread_group_cputime_adjusted() to get times for
1113 * the thread group, which consolidates times for all threads
1114 * in the group including the group leader.
1115 */
1116 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1117 spin_lock_irq(¤t->sighand->siglock);
1118 write_seqlock(&psig->stats_lock);
1119 psig->cutime += tgutime + sig->cutime;
1120 psig->cstime += tgstime + sig->cstime;
1121 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1122 psig->cmin_flt +=
1123 p->min_flt + sig->min_flt + sig->cmin_flt;
1124 psig->cmaj_flt +=
1125 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1126 psig->cnvcsw +=
1127 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1128 psig->cnivcsw +=
1129 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1130 psig->cinblock +=
1131 task_io_get_inblock(p) +
1132 sig->inblock + sig->cinblock;
1133 psig->coublock +=
1134 task_io_get_oublock(p) +
1135 sig->oublock + sig->coublock;
1136 maxrss = max(sig->maxrss, sig->cmaxrss);
1137 if (psig->cmaxrss < maxrss)
1138 psig->cmaxrss = maxrss;
1139 task_io_accounting_add(&psig->ioac, &p->ioac);
1140 task_io_accounting_add(&psig->ioac, &sig->ioac);
1141 write_sequnlock(&psig->stats_lock);
1142 spin_unlock_irq(¤t->sighand->siglock);
1143 }
1144
1145 retval = wo->wo_rusage
1146 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1147 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1148 ? p->signal->group_exit_code : p->exit_code;
1149 if (!retval && wo->wo_stat)
1150 retval = put_user(status, wo->wo_stat);
1151
1152 infop = wo->wo_info;
1153 if (!retval && infop)
1154 retval = put_user(SIGCHLD, &infop->si_signo);
1155 if (!retval && infop)
1156 retval = put_user(0, &infop->si_errno);
1157 if (!retval && infop) {
1158 int why;
1159
1160 if ((status & 0x7f) == 0) {
1161 why = CLD_EXITED;
1162 status >>= 8;
1163 } else {
1164 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1165 status &= 0x7f;
1166 }
1167 retval = put_user((short)why, &infop->si_code);
1168 if (!retval)
1169 retval = put_user(status, &infop->si_status);
1170 }
1171 if (!retval && infop)
1172 retval = put_user(pid, &infop->si_pid);
1173 if (!retval && infop)
1174 retval = put_user(uid, &infop->si_uid);
1175 if (!retval)
1176 retval = pid;
1177
1178 if (state == EXIT_TRACE) {
1179 write_lock_irq(&tasklist_lock);
1180 /* We dropped tasklist, ptracer could die and untrace */
1181 ptrace_unlink(p);
1182
1183 /* If parent wants a zombie, don't release it now */
1184 state = EXIT_ZOMBIE;
1185 if (do_notify_parent(p, p->exit_signal))
1186 state = EXIT_DEAD;
1187 p->exit_state = state;
1188 write_unlock_irq(&tasklist_lock);
1189 }
1190 if (state == EXIT_DEAD)
1191 release_task(p);
1192
1193 return retval;
1194}
1195
1196static int *task_stopped_code(struct task_struct *p, bool ptrace)
1197{
1198 if (ptrace) {
1199 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1200 return &p->exit_code;
1201 } else {
1202 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1203 return &p->signal->group_exit_code;
1204 }
1205 return NULL;
1206}
1207
1208/**
1209 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1210 * @wo: wait options
1211 * @ptrace: is the wait for ptrace
1212 * @p: task to wait for
1213 *
1214 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1215 *
1216 * CONTEXT:
1217 * read_lock(&tasklist_lock), which is released if return value is
1218 * non-zero. Also, grabs and releases @p->sighand->siglock.
1219 *
1220 * RETURNS:
1221 * 0 if wait condition didn't exist and search for other wait conditions
1222 * should continue. Non-zero return, -errno on failure and @p's pid on
1223 * success, implies that tasklist_lock is released and wait condition
1224 * search should terminate.
1225 */
1226static int wait_task_stopped(struct wait_opts *wo,
1227 int ptrace, struct task_struct *p)
1228{
1229 struct siginfo __user *infop;
1230 int retval, exit_code, *p_code, why;
1231 uid_t uid = 0; /* unneeded, required by compiler */
1232 pid_t pid;
1233
1234 /*
1235 * Traditionally we see ptrace'd stopped tasks regardless of options.
1236 */
1237 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1238 return 0;
1239
1240 if (!task_stopped_code(p, ptrace))
1241 return 0;
1242
1243 exit_code = 0;
1244 spin_lock_irq(&p->sighand->siglock);
1245
1246 p_code = task_stopped_code(p, ptrace);
1247 if (unlikely(!p_code))
1248 goto unlock_sig;
1249
1250 exit_code = *p_code;
1251 if (!exit_code)
1252 goto unlock_sig;
1253
1254 if (!unlikely(wo->wo_flags & WNOWAIT))
1255 *p_code = 0;
1256
1257 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1258unlock_sig:
1259 spin_unlock_irq(&p->sighand->siglock);
1260 if (!exit_code)
1261 return 0;
1262
1263 /*
1264 * Now we are pretty sure this task is interesting.
1265 * Make sure it doesn't get reaped out from under us while we
1266 * give up the lock and then examine it below. We don't want to
1267 * keep holding onto the tasklist_lock while we call getrusage and
1268 * possibly take page faults for user memory.
1269 */
1270 get_task_struct(p);
1271 pid = task_pid_vnr(p);
1272 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1273 read_unlock(&tasklist_lock);
1274 sched_annotate_sleep();
1275
1276 if (unlikely(wo->wo_flags & WNOWAIT))
1277 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1278
1279 retval = wo->wo_rusage
1280 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1281 if (!retval && wo->wo_stat)
1282 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1283
1284 infop = wo->wo_info;
1285 if (!retval && infop)
1286 retval = put_user(SIGCHLD, &infop->si_signo);
1287 if (!retval && infop)
1288 retval = put_user(0, &infop->si_errno);
1289 if (!retval && infop)
1290 retval = put_user((short)why, &infop->si_code);
1291 if (!retval && infop)
1292 retval = put_user(exit_code, &infop->si_status);
1293 if (!retval && infop)
1294 retval = put_user(pid, &infop->si_pid);
1295 if (!retval && infop)
1296 retval = put_user(uid, &infop->si_uid);
1297 if (!retval)
1298 retval = pid;
1299 put_task_struct(p);
1300
1301 BUG_ON(!retval);
1302 return retval;
1303}
1304
1305/*
1306 * Handle do_wait work for one task in a live, non-stopped state.
1307 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1308 * the lock and this task is uninteresting. If we return nonzero, we have
1309 * released the lock and the system call should return.
1310 */
1311static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1312{
1313 int retval;
1314 pid_t pid;
1315 uid_t uid;
1316
1317 if (!unlikely(wo->wo_flags & WCONTINUED))
1318 return 0;
1319
1320 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1321 return 0;
1322
1323 spin_lock_irq(&p->sighand->siglock);
1324 /* Re-check with the lock held. */
1325 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1326 spin_unlock_irq(&p->sighand->siglock);
1327 return 0;
1328 }
1329 if (!unlikely(wo->wo_flags & WNOWAIT))
1330 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1331 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1332 spin_unlock_irq(&p->sighand->siglock);
1333
1334 pid = task_pid_vnr(p);
1335 get_task_struct(p);
1336 read_unlock(&tasklist_lock);
1337 sched_annotate_sleep();
1338
1339 if (!wo->wo_info) {
1340 retval = wo->wo_rusage
1341 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1342 put_task_struct(p);
1343 if (!retval && wo->wo_stat)
1344 retval = put_user(0xffff, wo->wo_stat);
1345 if (!retval)
1346 retval = pid;
1347 } else {
1348 retval = wait_noreap_copyout(wo, p, pid, uid,
1349 CLD_CONTINUED, SIGCONT);
1350 BUG_ON(retval == 0);
1351 }
1352
1353 return retval;
1354}
1355
1356/*
1357 * Consider @p for a wait by @parent.
1358 *
1359 * -ECHILD should be in ->notask_error before the first call.
1360 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1361 * Returns zero if the search for a child should continue;
1362 * then ->notask_error is 0 if @p is an eligible child,
1363 * or another error from security_task_wait(), or still -ECHILD.
1364 */
1365static int wait_consider_task(struct wait_opts *wo, int ptrace,
1366 struct task_struct *p)
1367{
1368 /*
1369 * We can race with wait_task_zombie() from another thread.
1370 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1371 * can't confuse the checks below.
1372 */
1373 int exit_state = ACCESS_ONCE(p->exit_state);
1374 int ret;
1375
1376 if (unlikely(exit_state == EXIT_DEAD))
1377 return 0;
1378
1379 ret = eligible_child(wo, ptrace, p);
1380 if (!ret)
1381 return ret;
1382
1383 ret = security_task_wait(p);
1384 if (unlikely(ret < 0)) {
1385 /*
1386 * If we have not yet seen any eligible child,
1387 * then let this error code replace -ECHILD.
1388 * A permission error will give the user a clue
1389 * to look for security policy problems, rather
1390 * than for mysterious wait bugs.
1391 */
1392 if (wo->notask_error)
1393 wo->notask_error = ret;
1394 return 0;
1395 }
1396
1397 if (unlikely(exit_state == EXIT_TRACE)) {
1398 /*
1399 * ptrace == 0 means we are the natural parent. In this case
1400 * we should clear notask_error, debugger will notify us.
1401 */
1402 if (likely(!ptrace))
1403 wo->notask_error = 0;
1404 return 0;
1405 }
1406
1407 if (likely(!ptrace) && unlikely(p->ptrace)) {
1408 /*
1409 * If it is traced by its real parent's group, just pretend
1410 * the caller is ptrace_do_wait() and reap this child if it
1411 * is zombie.
1412 *
1413 * This also hides group stop state from real parent; otherwise
1414 * a single stop can be reported twice as group and ptrace stop.
1415 * If a ptracer wants to distinguish these two events for its
1416 * own children it should create a separate process which takes
1417 * the role of real parent.
1418 */
1419 if (!ptrace_reparented(p))
1420 ptrace = 1;
1421 }
1422
1423 /* slay zombie? */
1424 if (exit_state == EXIT_ZOMBIE) {
1425 /* we don't reap group leaders with subthreads */
1426 if (!delay_group_leader(p)) {
1427 /*
1428 * A zombie ptracee is only visible to its ptracer.
1429 * Notification and reaping will be cascaded to the
1430 * real parent when the ptracer detaches.
1431 */
1432 if (unlikely(ptrace) || likely(!p->ptrace))
1433 return wait_task_zombie(wo, p);
1434 }
1435
1436 /*
1437 * Allow access to stopped/continued state via zombie by
1438 * falling through. Clearing of notask_error is complex.
1439 *
1440 * When !@ptrace:
1441 *
1442 * If WEXITED is set, notask_error should naturally be
1443 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1444 * so, if there are live subthreads, there are events to
1445 * wait for. If all subthreads are dead, it's still safe
1446 * to clear - this function will be called again in finite
1447 * amount time once all the subthreads are released and
1448 * will then return without clearing.
1449 *
1450 * When @ptrace:
1451 *
1452 * Stopped state is per-task and thus can't change once the
1453 * target task dies. Only continued and exited can happen.
1454 * Clear notask_error if WCONTINUED | WEXITED.
1455 */
1456 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1457 wo->notask_error = 0;
1458 } else {
1459 /*
1460 * @p is alive and it's gonna stop, continue or exit, so
1461 * there always is something to wait for.
1462 */
1463 wo->notask_error = 0;
1464 }
1465
1466 /*
1467 * Wait for stopped. Depending on @ptrace, different stopped state
1468 * is used and the two don't interact with each other.
1469 */
1470 ret = wait_task_stopped(wo, ptrace, p);
1471 if (ret)
1472 return ret;
1473
1474 /*
1475 * Wait for continued. There's only one continued state and the
1476 * ptracer can consume it which can confuse the real parent. Don't
1477 * use WCONTINUED from ptracer. You don't need or want it.
1478 */
1479 return wait_task_continued(wo, p);
1480}
1481
1482/*
1483 * Do the work of do_wait() for one thread in the group, @tsk.
1484 *
1485 * -ECHILD should be in ->notask_error before the first call.
1486 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1487 * Returns zero if the search for a child should continue; then
1488 * ->notask_error is 0 if there were any eligible children,
1489 * or another error from security_task_wait(), or still -ECHILD.
1490 */
1491static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1492{
1493 struct task_struct *p;
1494
1495 list_for_each_entry(p, &tsk->children, sibling) {
1496 int ret = wait_consider_task(wo, 0, p);
1497
1498 if (ret)
1499 return ret;
1500 }
1501
1502 return 0;
1503}
1504
1505static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1506{
1507 struct task_struct *p;
1508
1509 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1510 int ret = wait_consider_task(wo, 1, p);
1511
1512 if (ret)
1513 return ret;
1514 }
1515
1516 return 0;
1517}
1518
1519static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1520 int sync, void *key)
1521{
1522 struct wait_opts *wo = container_of(wait, struct wait_opts,
1523 child_wait);
1524 struct task_struct *p = key;
1525
1526 if (!eligible_pid(wo, p))
1527 return 0;
1528
1529 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1530 return 0;
1531
1532 return default_wake_function(wait, mode, sync, key);
1533}
1534
1535void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1536{
1537 __wake_up_sync_key(&parent->signal->wait_chldexit,
1538 TASK_INTERRUPTIBLE, 1, p);
1539}
1540
1541static long do_wait(struct wait_opts *wo)
1542{
1543 struct task_struct *tsk;
1544 int retval;
1545
1546 trace_sched_process_wait(wo->wo_pid);
1547
1548 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1549 wo->child_wait.private = current;
1550 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1551repeat:
1552 /*
1553 * If there is nothing that can match our criteria, just get out.
1554 * We will clear ->notask_error to zero if we see any child that
1555 * might later match our criteria, even if we are not able to reap
1556 * it yet.
1557 */
1558 wo->notask_error = -ECHILD;
1559 if ((wo->wo_type < PIDTYPE_MAX) &&
1560 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1561 goto notask;
1562
1563 set_current_state(TASK_INTERRUPTIBLE);
1564 read_lock(&tasklist_lock);
1565 tsk = current;
1566 do {
1567 retval = do_wait_thread(wo, tsk);
1568 if (retval)
1569 goto end;
1570
1571 retval = ptrace_do_wait(wo, tsk);
1572 if (retval)
1573 goto end;
1574
1575 if (wo->wo_flags & __WNOTHREAD)
1576 break;
1577 } while_each_thread(current, tsk);
1578 read_unlock(&tasklist_lock);
1579
1580notask:
1581 retval = wo->notask_error;
1582 if (!retval && !(wo->wo_flags & WNOHANG)) {
1583 retval = -ERESTARTSYS;
1584 if (!signal_pending(current)) {
1585 schedule();
1586 goto repeat;
1587 }
1588 }
1589end:
1590 __set_current_state(TASK_RUNNING);
1591 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1592 return retval;
1593}
1594
1595SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1596 infop, int, options, struct rusage __user *, ru)
1597{
1598 struct wait_opts wo;
1599 struct pid *pid = NULL;
1600 enum pid_type type;
1601 long ret;
1602
1603 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1604 __WNOTHREAD|__WCLONE|__WALL))
1605 return -EINVAL;
1606 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1607 return -EINVAL;
1608
1609 switch (which) {
1610 case P_ALL:
1611 type = PIDTYPE_MAX;
1612 break;
1613 case P_PID:
1614 type = PIDTYPE_PID;
1615 if (upid <= 0)
1616 return -EINVAL;
1617 break;
1618 case P_PGID:
1619 type = PIDTYPE_PGID;
1620 if (upid <= 0)
1621 return -EINVAL;
1622 break;
1623 default:
1624 return -EINVAL;
1625 }
1626
1627 if (type < PIDTYPE_MAX)
1628 pid = find_get_pid(upid);
1629
1630 wo.wo_type = type;
1631 wo.wo_pid = pid;
1632 wo.wo_flags = options;
1633 wo.wo_info = infop;
1634 wo.wo_stat = NULL;
1635 wo.wo_rusage = ru;
1636 ret = do_wait(&wo);
1637
1638 if (ret > 0) {
1639 ret = 0;
1640 } else if (infop) {
1641 /*
1642 * For a WNOHANG return, clear out all the fields
1643 * we would set so the user can easily tell the
1644 * difference.
1645 */
1646 if (!ret)
1647 ret = put_user(0, &infop->si_signo);
1648 if (!ret)
1649 ret = put_user(0, &infop->si_errno);
1650 if (!ret)
1651 ret = put_user(0, &infop->si_code);
1652 if (!ret)
1653 ret = put_user(0, &infop->si_pid);
1654 if (!ret)
1655 ret = put_user(0, &infop->si_uid);
1656 if (!ret)
1657 ret = put_user(0, &infop->si_status);
1658 }
1659
1660 put_pid(pid);
1661 return ret;
1662}
1663
1664SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1665 int, options, struct rusage __user *, ru)
1666{
1667 struct wait_opts wo;
1668 struct pid *pid = NULL;
1669 enum pid_type type;
1670 long ret;
1671
1672 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1673 __WNOTHREAD|__WCLONE|__WALL))
1674 return -EINVAL;
1675
1676 if (upid == -1)
1677 type = PIDTYPE_MAX;
1678 else if (upid < 0) {
1679 type = PIDTYPE_PGID;
1680 pid = find_get_pid(-upid);
1681 } else if (upid == 0) {
1682 type = PIDTYPE_PGID;
1683 pid = get_task_pid(current, PIDTYPE_PGID);
1684 } else /* upid > 0 */ {
1685 type = PIDTYPE_PID;
1686 pid = find_get_pid(upid);
1687 }
1688
1689 wo.wo_type = type;
1690 wo.wo_pid = pid;
1691 wo.wo_flags = options | WEXITED;
1692 wo.wo_info = NULL;
1693 wo.wo_stat = stat_addr;
1694 wo.wo_rusage = ru;
1695 ret = do_wait(&wo);
1696 put_pid(pid);
1697
1698 return ret;
1699}
1700
1701#ifdef __ARCH_WANT_SYS_WAITPID
1702
1703/*
1704 * sys_waitpid() remains for compatibility. waitpid() should be
1705 * implemented by calling sys_wait4() from libc.a.
1706 */
1707SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1708{
1709 return sys_wait4(pid, stat_addr, options, NULL);
1710}
1711
1712#endif
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/exit.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/mm.h>
9#include <linux/slab.h>
10#include <linux/sched/autogroup.h>
11#include <linux/sched/mm.h>
12#include <linux/sched/stat.h>
13#include <linux/sched/task.h>
14#include <linux/sched/task_stack.h>
15#include <linux/sched/cputime.h>
16#include <linux/interrupt.h>
17#include <linux/module.h>
18#include <linux/capability.h>
19#include <linux/completion.h>
20#include <linux/personality.h>
21#include <linux/tty.h>
22#include <linux/iocontext.h>
23#include <linux/key.h>
24#include <linux/cpu.h>
25#include <linux/acct.h>
26#include <linux/tsacct_kern.h>
27#include <linux/file.h>
28#include <linux/fdtable.h>
29#include <linux/freezer.h>
30#include <linux/binfmts.h>
31#include <linux/nsproxy.h>
32#include <linux/pid_namespace.h>
33#include <linux/ptrace.h>
34#include <linux/profile.h>
35#include <linux/mount.h>
36#include <linux/proc_fs.h>
37#include <linux/kthread.h>
38#include <linux/mempolicy.h>
39#include <linux/taskstats_kern.h>
40#include <linux/delayacct.h>
41#include <linux/cgroup.h>
42#include <linux/syscalls.h>
43#include <linux/signal.h>
44#include <linux/posix-timers.h>
45#include <linux/cn_proc.h>
46#include <linux/mutex.h>
47#include <linux/futex.h>
48#include <linux/pipe_fs_i.h>
49#include <linux/audit.h> /* for audit_free() */
50#include <linux/resource.h>
51#include <linux/blkdev.h>
52#include <linux/task_io_accounting_ops.h>
53#include <linux/tracehook.h>
54#include <linux/fs_struct.h>
55#include <linux/init_task.h>
56#include <linux/perf_event.h>
57#include <trace/events/sched.h>
58#include <linux/hw_breakpoint.h>
59#include <linux/oom.h>
60#include <linux/writeback.h>
61#include <linux/shm.h>
62#include <linux/kcov.h>
63#include <linux/random.h>
64#include <linux/rcuwait.h>
65#include <linux/compat.h>
66
67#include <linux/uaccess.h>
68#include <asm/unistd.h>
69#include <asm/pgtable.h>
70#include <asm/mmu_context.h>
71
72static 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 */
92static void __exit_signal(struct task_struct *tsk)
93{
94 struct signal_struct *sig = tsk->signal;
95 bool group_dead = thread_group_leader(tsk);
96 struct sighand_struct *sighand;
97 struct tty_struct *uninitialized_var(tty);
98 u64 utime, stime;
99
100 sighand = rcu_dereference_check(tsk->sighand,
101 lockdep_tasklist_lock_is_held());
102 spin_lock(&sighand->siglock);
103
104#ifdef CONFIG_POSIX_TIMERS
105 posix_cpu_timers_exit(tsk);
106 if (group_dead) {
107 posix_cpu_timers_exit_group(tsk);
108 } else {
109 /*
110 * This can only happen if the caller is de_thread().
111 * FIXME: this is the temporary hack, we should teach
112 * posix-cpu-timers to handle this case correctly.
113 */
114 if (unlikely(has_group_leader_pid(tsk)))
115 posix_cpu_timers_exit_group(tsk);
116 }
117#endif
118
119 if (group_dead) {
120 tty = sig->tty;
121 sig->tty = NULL;
122 } else {
123 /*
124 * If there is any task waiting for the group exit
125 * then notify it:
126 */
127 if (sig->notify_count > 0 && !--sig->notify_count)
128 wake_up_process(sig->group_exit_task);
129
130 if (tsk == sig->curr_target)
131 sig->curr_target = next_thread(tsk);
132 }
133
134 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
135 sizeof(unsigned long long));
136
137 /*
138 * Accumulate here the counters for all threads as they die. We could
139 * skip the group leader because it is the last user of signal_struct,
140 * but we want to avoid the race with thread_group_cputime() which can
141 * see the empty ->thread_head list.
142 */
143 task_cputime(tsk, &utime, &stime);
144 write_seqlock(&sig->stats_lock);
145 sig->utime += utime;
146 sig->stime += stime;
147 sig->gtime += task_gtime(tsk);
148 sig->min_flt += tsk->min_flt;
149 sig->maj_flt += tsk->maj_flt;
150 sig->nvcsw += tsk->nvcsw;
151 sig->nivcsw += tsk->nivcsw;
152 sig->inblock += task_io_get_inblock(tsk);
153 sig->oublock += task_io_get_oublock(tsk);
154 task_io_accounting_add(&sig->ioac, &tsk->ioac);
155 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
156 sig->nr_threads--;
157 __unhash_process(tsk, group_dead);
158 write_sequnlock(&sig->stats_lock);
159
160 /*
161 * Do this under ->siglock, we can race with another thread
162 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
163 */
164 flush_sigqueue(&tsk->pending);
165 tsk->sighand = NULL;
166 spin_unlock(&sighand->siglock);
167
168 __cleanup_sighand(sighand);
169 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
170 if (group_dead) {
171 flush_sigqueue(&sig->shared_pending);
172 tty_kref_put(tty);
173 }
174}
175
176static void delayed_put_task_struct(struct rcu_head *rhp)
177{
178 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
179
180 perf_event_delayed_put(tsk);
181 trace_sched_process_free(tsk);
182 put_task_struct(tsk);
183}
184
185void put_task_struct_rcu_user(struct task_struct *task)
186{
187 if (refcount_dec_and_test(&task->rcu_users))
188 call_rcu(&task->rcu, delayed_put_task_struct);
189}
190
191void release_task(struct task_struct *p)
192{
193 struct task_struct *leader;
194 int zap_leader;
195repeat:
196 /* don't need to get the RCU readlock here - the process is dead and
197 * can't be modifying its own credentials. But shut RCU-lockdep up */
198 rcu_read_lock();
199 atomic_dec(&__task_cred(p)->user->processes);
200 rcu_read_unlock();
201
202 proc_flush_task(p);
203 cgroup_release(p);
204
205 write_lock_irq(&tasklist_lock);
206 ptrace_release_task(p);
207 __exit_signal(p);
208
209 /*
210 * If we are the last non-leader member of the thread
211 * group, and the leader is zombie, then notify the
212 * group leader's parent process. (if it wants notification.)
213 */
214 zap_leader = 0;
215 leader = p->group_leader;
216 if (leader != p && thread_group_empty(leader)
217 && leader->exit_state == EXIT_ZOMBIE) {
218 /*
219 * If we were the last child thread and the leader has
220 * exited already, and the leader's parent ignores SIGCHLD,
221 * then we are the one who should release the leader.
222 */
223 zap_leader = do_notify_parent(leader, leader->exit_signal);
224 if (zap_leader)
225 leader->exit_state = EXIT_DEAD;
226 }
227
228 write_unlock_irq(&tasklist_lock);
229 release_thread(p);
230 put_task_struct_rcu_user(p);
231
232 p = leader;
233 if (unlikely(zap_leader))
234 goto repeat;
235}
236
237void rcuwait_wake_up(struct rcuwait *w)
238{
239 struct task_struct *task;
240
241 rcu_read_lock();
242
243 /*
244 * Order condition vs @task, such that everything prior to the load
245 * of @task is visible. This is the condition as to why the user called
246 * rcuwait_trywake() in the first place. Pairs with set_current_state()
247 * barrier (A) in rcuwait_wait_event().
248 *
249 * WAIT WAKE
250 * [S] tsk = current [S] cond = true
251 * MB (A) MB (B)
252 * [L] cond [L] tsk
253 */
254 smp_mb(); /* (B) */
255
256 task = rcu_dereference(w->task);
257 if (task)
258 wake_up_process(task);
259 rcu_read_unlock();
260}
261
262/*
263 * Determine if a process group is "orphaned", according to the POSIX
264 * definition in 2.2.2.52. Orphaned process groups are not to be affected
265 * by terminal-generated stop signals. Newly orphaned process groups are
266 * to receive a SIGHUP and a SIGCONT.
267 *
268 * "I ask you, have you ever known what it is to be an orphan?"
269 */
270static 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
289int 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
300static 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 */
317static void
318kill_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 */
347void mm_update_next_owner(struct mm_struct *mm)
348{
349 struct task_struct *c, *g, *p = current;
350
351retry:
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
407assign_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 */
435static void exit_mm(void)
436{
437 struct mm_struct *mm = current->mm;
438 struct core_state *core_state;
439
440 mm_release(current, mm);
441 if (!mm)
442 return;
443 sync_mm_rss(mm);
444 /*
445 * Serialize with any possible pending coredump.
446 * We must hold mmap_sem around checking core_state
447 * and clearing tsk->mm. The core-inducing thread
448 * will increment ->nr_threads for each thread in the
449 * group with ->mm != NULL.
450 */
451 down_read(&mm->mmap_sem);
452 core_state = mm->core_state;
453 if (core_state) {
454 struct core_thread self;
455
456 up_read(&mm->mmap_sem);
457
458 self.task = current;
459 self.next = xchg(&core_state->dumper.next, &self);
460 /*
461 * Implies mb(), the result of xchg() must be visible
462 * to core_state->dumper.
463 */
464 if (atomic_dec_and_test(&core_state->nr_threads))
465 complete(&core_state->startup);
466
467 for (;;) {
468 set_current_state(TASK_UNINTERRUPTIBLE);
469 if (!self.task) /* see coredump_finish() */
470 break;
471 freezable_schedule();
472 }
473 __set_current_state(TASK_RUNNING);
474 down_read(&mm->mmap_sem);
475 }
476 mmgrab(mm);
477 BUG_ON(mm != current->active_mm);
478 /* more a memory barrier than a real lock */
479 task_lock(current);
480 current->mm = NULL;
481 up_read(&mm->mmap_sem);
482 enter_lazy_tlb(mm, current);
483 task_unlock(current);
484 mm_update_next_owner(mm);
485 mmput(mm);
486 if (test_thread_flag(TIF_MEMDIE))
487 exit_oom_victim();
488}
489
490static struct task_struct *find_alive_thread(struct task_struct *p)
491{
492 struct task_struct *t;
493
494 for_each_thread(p, t) {
495 if (!(t->flags & PF_EXITING))
496 return t;
497 }
498 return NULL;
499}
500
501static struct task_struct *find_child_reaper(struct task_struct *father,
502 struct list_head *dead)
503 __releases(&tasklist_lock)
504 __acquires(&tasklist_lock)
505{
506 struct pid_namespace *pid_ns = task_active_pid_ns(father);
507 struct task_struct *reaper = pid_ns->child_reaper;
508 struct task_struct *p, *n;
509
510 if (likely(reaper != father))
511 return reaper;
512
513 reaper = find_alive_thread(father);
514 if (reaper) {
515 pid_ns->child_reaper = reaper;
516 return reaper;
517 }
518
519 write_unlock_irq(&tasklist_lock);
520 if (unlikely(pid_ns == &init_pid_ns)) {
521 panic("Attempted to kill init! exitcode=0x%08x\n",
522 father->signal->group_exit_code ?: father->exit_code);
523 }
524
525 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
526 list_del_init(&p->ptrace_entry);
527 release_task(p);
528 }
529
530 zap_pid_ns_processes(pid_ns);
531 write_lock_irq(&tasklist_lock);
532
533 return father;
534}
535
536/*
537 * When we die, we re-parent all our children, and try to:
538 * 1. give them to another thread in our thread group, if such a member exists
539 * 2. give it to the first ancestor process which prctl'd itself as a
540 * child_subreaper for its children (like a service manager)
541 * 3. give it to the init process (PID 1) in our pid namespace
542 */
543static struct task_struct *find_new_reaper(struct task_struct *father,
544 struct task_struct *child_reaper)
545{
546 struct task_struct *thread, *reaper;
547
548 thread = find_alive_thread(father);
549 if (thread)
550 return thread;
551
552 if (father->signal->has_child_subreaper) {
553 unsigned int ns_level = task_pid(father)->level;
554 /*
555 * Find the first ->is_child_subreaper ancestor in our pid_ns.
556 * We can't check reaper != child_reaper to ensure we do not
557 * cross the namespaces, the exiting parent could be injected
558 * by setns() + fork().
559 * We check pid->level, this is slightly more efficient than
560 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
561 */
562 for (reaper = father->real_parent;
563 task_pid(reaper)->level == ns_level;
564 reaper = reaper->real_parent) {
565 if (reaper == &init_task)
566 break;
567 if (!reaper->signal->is_child_subreaper)
568 continue;
569 thread = find_alive_thread(reaper);
570 if (thread)
571 return thread;
572 }
573 }
574
575 return child_reaper;
576}
577
578/*
579* Any that need to be release_task'd are put on the @dead list.
580 */
581static void reparent_leader(struct task_struct *father, struct task_struct *p,
582 struct list_head *dead)
583{
584 if (unlikely(p->exit_state == EXIT_DEAD))
585 return;
586
587 /* We don't want people slaying init. */
588 p->exit_signal = SIGCHLD;
589
590 /* If it has exited notify the new parent about this child's death. */
591 if (!p->ptrace &&
592 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
593 if (do_notify_parent(p, p->exit_signal)) {
594 p->exit_state = EXIT_DEAD;
595 list_add(&p->ptrace_entry, dead);
596 }
597 }
598
599 kill_orphaned_pgrp(p, father);
600}
601
602/*
603 * This does two things:
604 *
605 * A. Make init inherit all the child processes
606 * B. Check to see if any process groups have become orphaned
607 * as a result of our exiting, and if they have any stopped
608 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
609 */
610static void forget_original_parent(struct task_struct *father,
611 struct list_head *dead)
612{
613 struct task_struct *p, *t, *reaper;
614
615 if (unlikely(!list_empty(&father->ptraced)))
616 exit_ptrace(father, dead);
617
618 /* Can drop and reacquire tasklist_lock */
619 reaper = find_child_reaper(father, dead);
620 if (list_empty(&father->children))
621 return;
622
623 reaper = find_new_reaper(father, reaper);
624 list_for_each_entry(p, &father->children, sibling) {
625 for_each_thread(p, t) {
626 t->real_parent = reaper;
627 BUG_ON((!t->ptrace) != (t->parent == father));
628 if (likely(!t->ptrace))
629 t->parent = t->real_parent;
630 if (t->pdeath_signal)
631 group_send_sig_info(t->pdeath_signal,
632 SEND_SIG_NOINFO, t,
633 PIDTYPE_TGID);
634 }
635 /*
636 * If this is a threaded reparent there is no need to
637 * notify anyone anything has happened.
638 */
639 if (!same_thread_group(reaper, father))
640 reparent_leader(father, p, dead);
641 }
642 list_splice_tail_init(&father->children, &reaper->children);
643}
644
645/*
646 * Send signals to all our closest relatives so that they know
647 * to properly mourn us..
648 */
649static void exit_notify(struct task_struct *tsk, int group_dead)
650{
651 bool autoreap;
652 struct task_struct *p, *n;
653 LIST_HEAD(dead);
654
655 write_lock_irq(&tasklist_lock);
656 forget_original_parent(tsk, &dead);
657
658 if (group_dead)
659 kill_orphaned_pgrp(tsk->group_leader, NULL);
660
661 tsk->exit_state = EXIT_ZOMBIE;
662 if (unlikely(tsk->ptrace)) {
663 int sig = thread_group_leader(tsk) &&
664 thread_group_empty(tsk) &&
665 !ptrace_reparented(tsk) ?
666 tsk->exit_signal : SIGCHLD;
667 autoreap = do_notify_parent(tsk, sig);
668 } else if (thread_group_leader(tsk)) {
669 autoreap = thread_group_empty(tsk) &&
670 do_notify_parent(tsk, tsk->exit_signal);
671 } else {
672 autoreap = true;
673 }
674
675 if (autoreap) {
676 tsk->exit_state = EXIT_DEAD;
677 list_add(&tsk->ptrace_entry, &dead);
678 }
679
680 /* mt-exec, de_thread() is waiting for group leader */
681 if (unlikely(tsk->signal->notify_count < 0))
682 wake_up_process(tsk->signal->group_exit_task);
683 write_unlock_irq(&tasklist_lock);
684
685 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
686 list_del_init(&p->ptrace_entry);
687 release_task(p);
688 }
689}
690
691#ifdef CONFIG_DEBUG_STACK_USAGE
692static void check_stack_usage(void)
693{
694 static DEFINE_SPINLOCK(low_water_lock);
695 static int lowest_to_date = THREAD_SIZE;
696 unsigned long free;
697
698 free = stack_not_used(current);
699
700 if (free >= lowest_to_date)
701 return;
702
703 spin_lock(&low_water_lock);
704 if (free < lowest_to_date) {
705 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
706 current->comm, task_pid_nr(current), free);
707 lowest_to_date = free;
708 }
709 spin_unlock(&low_water_lock);
710}
711#else
712static inline void check_stack_usage(void) {}
713#endif
714
715void __noreturn do_exit(long code)
716{
717 struct task_struct *tsk = current;
718 int group_dead;
719
720 profile_task_exit(tsk);
721 kcov_task_exit(tsk);
722
723 WARN_ON(blk_needs_flush_plug(tsk));
724
725 if (unlikely(in_interrupt()))
726 panic("Aiee, killing interrupt handler!");
727 if (unlikely(!tsk->pid))
728 panic("Attempted to kill the idle task!");
729
730 /*
731 * If do_exit is called because this processes oopsed, it's possible
732 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
733 * continuing. Amongst other possible reasons, this is to prevent
734 * mm_release()->clear_child_tid() from writing to a user-controlled
735 * kernel address.
736 */
737 set_fs(USER_DS);
738
739 ptrace_event(PTRACE_EVENT_EXIT, code);
740
741 validate_creds_for_do_exit(tsk);
742
743 /*
744 * We're taking recursive faults here in do_exit. Safest is to just
745 * leave this task alone and wait for reboot.
746 */
747 if (unlikely(tsk->flags & PF_EXITING)) {
748 pr_alert("Fixing recursive fault but reboot is needed!\n");
749 /*
750 * We can do this unlocked here. The futex code uses
751 * this flag just to verify whether the pi state
752 * cleanup has been done or not. In the worst case it
753 * loops once more. We pretend that the cleanup was
754 * done as there is no way to return. Either the
755 * OWNER_DIED bit is set by now or we push the blocked
756 * task into the wait for ever nirwana as well.
757 */
758 tsk->flags |= PF_EXITPIDONE;
759 set_current_state(TASK_UNINTERRUPTIBLE);
760 schedule();
761 }
762
763 exit_signals(tsk); /* sets PF_EXITING */
764 /*
765 * Ensure that all new tsk->pi_lock acquisitions must observe
766 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
767 */
768 smp_mb();
769 /*
770 * Ensure that we must observe the pi_state in exit_mm() ->
771 * mm_release() -> exit_pi_state_list().
772 */
773 raw_spin_lock_irq(&tsk->pi_lock);
774 raw_spin_unlock_irq(&tsk->pi_lock);
775
776 if (unlikely(in_atomic())) {
777 pr_info("note: %s[%d] exited with preempt_count %d\n",
778 current->comm, task_pid_nr(current),
779 preempt_count());
780 preempt_count_set(PREEMPT_ENABLED);
781 }
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#ifdef CONFIG_POSIX_TIMERS
790 hrtimer_cancel(&tsk->signal->real_timer);
791 exit_itimers(tsk->signal);
792#endif
793 if (tsk->mm)
794 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
795 }
796 acct_collect(code, group_dead);
797 if (group_dead)
798 tty_audit_exit();
799 audit_free(tsk);
800
801 tsk->exit_code = code;
802 taskstats_exit(tsk, group_dead);
803
804 exit_mm();
805
806 if (group_dead)
807 acct_process();
808 trace_sched_process_exit(tsk);
809
810 exit_sem(tsk);
811 exit_shm(tsk);
812 exit_files(tsk);
813 exit_fs(tsk);
814 if (group_dead)
815 disassociate_ctty(1);
816 exit_task_namespaces(tsk);
817 exit_task_work(tsk);
818 exit_thread(tsk);
819 exit_umh(tsk);
820
821 /*
822 * Flush inherited counters to the parent - before the parent
823 * gets woken up by child-exit notifications.
824 *
825 * because of cgroup mode, must be called before cgroup_exit()
826 */
827 perf_event_exit_task(tsk);
828
829 sched_autogroup_exit_task(tsk);
830 cgroup_exit(tsk);
831
832 /*
833 * FIXME: do that only when needed, using sched_exit tracepoint
834 */
835 flush_ptrace_hw_breakpoint(tsk);
836
837 exit_tasks_rcu_start();
838 exit_notify(tsk, group_dead);
839 proc_exit_connector(tsk);
840 mpol_put_task_policy(tsk);
841#ifdef CONFIG_FUTEX
842 if (unlikely(current->pi_state_cache))
843 kfree(current->pi_state_cache);
844#endif
845 /*
846 * Make sure we are holding no locks:
847 */
848 debug_check_no_locks_held();
849 /*
850 * We can do this unlocked here. The futex code uses this flag
851 * just to verify whether the pi state cleanup has been done
852 * or not. In the worst case it loops once more.
853 */
854 tsk->flags |= PF_EXITPIDONE;
855
856 if (tsk->io_context)
857 exit_io_context(tsk);
858
859 if (tsk->splice_pipe)
860 free_pipe_info(tsk->splice_pipe);
861
862 if (tsk->task_frag.page)
863 put_page(tsk->task_frag.page);
864
865 validate_creds_for_do_exit(tsk);
866
867 check_stack_usage();
868 preempt_disable();
869 if (tsk->nr_dirtied)
870 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
871 exit_rcu();
872 exit_tasks_rcu_finish();
873
874 lockdep_free_task(tsk);
875 do_task_dead();
876}
877EXPORT_SYMBOL_GPL(do_exit);
878
879void complete_and_exit(struct completion *comp, long code)
880{
881 if (comp)
882 complete(comp);
883
884 do_exit(code);
885}
886EXPORT_SYMBOL(complete_and_exit);
887
888SYSCALL_DEFINE1(exit, int, error_code)
889{
890 do_exit((error_code&0xff)<<8);
891}
892
893/*
894 * Take down every thread in the group. This is called by fatal signals
895 * as well as by sys_exit_group (below).
896 */
897void
898do_group_exit(int exit_code)
899{
900 struct signal_struct *sig = current->signal;
901
902 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
903
904 if (signal_group_exit(sig))
905 exit_code = sig->group_exit_code;
906 else if (!thread_group_empty(current)) {
907 struct sighand_struct *const sighand = current->sighand;
908
909 spin_lock_irq(&sighand->siglock);
910 if (signal_group_exit(sig))
911 /* Another thread got here before we took the lock. */
912 exit_code = sig->group_exit_code;
913 else {
914 sig->group_exit_code = exit_code;
915 sig->flags = SIGNAL_GROUP_EXIT;
916 zap_other_threads(current);
917 }
918 spin_unlock_irq(&sighand->siglock);
919 }
920
921 do_exit(exit_code);
922 /* NOTREACHED */
923}
924
925/*
926 * this kills every thread in the thread group. Note that any externally
927 * wait4()-ing process will get the correct exit code - even if this
928 * thread is not the thread group leader.
929 */
930SYSCALL_DEFINE1(exit_group, int, error_code)
931{
932 do_group_exit((error_code & 0xff) << 8);
933 /* NOTREACHED */
934 return 0;
935}
936
937struct waitid_info {
938 pid_t pid;
939 uid_t uid;
940 int status;
941 int cause;
942};
943
944struct wait_opts {
945 enum pid_type wo_type;
946 int wo_flags;
947 struct pid *wo_pid;
948
949 struct waitid_info *wo_info;
950 int wo_stat;
951 struct rusage *wo_rusage;
952
953 wait_queue_entry_t child_wait;
954 int notask_error;
955};
956
957static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
958{
959 return wo->wo_type == PIDTYPE_MAX ||
960 task_pid_type(p, wo->wo_type) == wo->wo_pid;
961}
962
963static int
964eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
965{
966 if (!eligible_pid(wo, p))
967 return 0;
968
969 /*
970 * Wait for all children (clone and not) if __WALL is set or
971 * if it is traced by us.
972 */
973 if (ptrace || (wo->wo_flags & __WALL))
974 return 1;
975
976 /*
977 * Otherwise, wait for clone children *only* if __WCLONE is set;
978 * otherwise, wait for non-clone children *only*.
979 *
980 * Note: a "clone" child here is one that reports to its parent
981 * using a signal other than SIGCHLD, or a non-leader thread which
982 * we can only see if it is traced by us.
983 */
984 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
985 return 0;
986
987 return 1;
988}
989
990/*
991 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
992 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
993 * the lock and this task is uninteresting. If we return nonzero, we have
994 * released the lock and the system call should return.
995 */
996static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
997{
998 int state, status;
999 pid_t pid = task_pid_vnr(p);
1000 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1001 struct waitid_info *infop;
1002
1003 if (!likely(wo->wo_flags & WEXITED))
1004 return 0;
1005
1006 if (unlikely(wo->wo_flags & WNOWAIT)) {
1007 status = p->exit_code;
1008 get_task_struct(p);
1009 read_unlock(&tasklist_lock);
1010 sched_annotate_sleep();
1011 if (wo->wo_rusage)
1012 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1013 put_task_struct(p);
1014 goto out_info;
1015 }
1016 /*
1017 * Move the task's state to DEAD/TRACE, only one thread can do this.
1018 */
1019 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1020 EXIT_TRACE : EXIT_DEAD;
1021 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1022 return 0;
1023 /*
1024 * We own this thread, nobody else can reap it.
1025 */
1026 read_unlock(&tasklist_lock);
1027 sched_annotate_sleep();
1028
1029 /*
1030 * Check thread_group_leader() to exclude the traced sub-threads.
1031 */
1032 if (state == EXIT_DEAD && thread_group_leader(p)) {
1033 struct signal_struct *sig = p->signal;
1034 struct signal_struct *psig = current->signal;
1035 unsigned long maxrss;
1036 u64 tgutime, tgstime;
1037
1038 /*
1039 * The resource counters for the group leader are in its
1040 * own task_struct. Those for dead threads in the group
1041 * are in its signal_struct, as are those for the child
1042 * processes it has previously reaped. All these
1043 * accumulate in the parent's signal_struct c* fields.
1044 *
1045 * We don't bother to take a lock here to protect these
1046 * p->signal fields because the whole thread group is dead
1047 * and nobody can change them.
1048 *
1049 * psig->stats_lock also protects us from our sub-theads
1050 * which can reap other children at the same time. Until
1051 * we change k_getrusage()-like users to rely on this lock
1052 * we have to take ->siglock as well.
1053 *
1054 * We use thread_group_cputime_adjusted() to get times for
1055 * the thread group, which consolidates times for all threads
1056 * in the group including the group leader.
1057 */
1058 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1059 spin_lock_irq(¤t->sighand->siglock);
1060 write_seqlock(&psig->stats_lock);
1061 psig->cutime += tgutime + sig->cutime;
1062 psig->cstime += tgstime + sig->cstime;
1063 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1064 psig->cmin_flt +=
1065 p->min_flt + sig->min_flt + sig->cmin_flt;
1066 psig->cmaj_flt +=
1067 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1068 psig->cnvcsw +=
1069 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1070 psig->cnivcsw +=
1071 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1072 psig->cinblock +=
1073 task_io_get_inblock(p) +
1074 sig->inblock + sig->cinblock;
1075 psig->coublock +=
1076 task_io_get_oublock(p) +
1077 sig->oublock + sig->coublock;
1078 maxrss = max(sig->maxrss, sig->cmaxrss);
1079 if (psig->cmaxrss < maxrss)
1080 psig->cmaxrss = maxrss;
1081 task_io_accounting_add(&psig->ioac, &p->ioac);
1082 task_io_accounting_add(&psig->ioac, &sig->ioac);
1083 write_sequnlock(&psig->stats_lock);
1084 spin_unlock_irq(¤t->sighand->siglock);
1085 }
1086
1087 if (wo->wo_rusage)
1088 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1089 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1090 ? p->signal->group_exit_code : p->exit_code;
1091 wo->wo_stat = status;
1092
1093 if (state == EXIT_TRACE) {
1094 write_lock_irq(&tasklist_lock);
1095 /* We dropped tasklist, ptracer could die and untrace */
1096 ptrace_unlink(p);
1097
1098 /* If parent wants a zombie, don't release it now */
1099 state = EXIT_ZOMBIE;
1100 if (do_notify_parent(p, p->exit_signal))
1101 state = EXIT_DEAD;
1102 p->exit_state = state;
1103 write_unlock_irq(&tasklist_lock);
1104 }
1105 if (state == EXIT_DEAD)
1106 release_task(p);
1107
1108out_info:
1109 infop = wo->wo_info;
1110 if (infop) {
1111 if ((status & 0x7f) == 0) {
1112 infop->cause = CLD_EXITED;
1113 infop->status = status >> 8;
1114 } else {
1115 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1116 infop->status = status & 0x7f;
1117 }
1118 infop->pid = pid;
1119 infop->uid = uid;
1120 }
1121
1122 return pid;
1123}
1124
1125static int *task_stopped_code(struct task_struct *p, bool ptrace)
1126{
1127 if (ptrace) {
1128 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1129 return &p->exit_code;
1130 } else {
1131 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1132 return &p->signal->group_exit_code;
1133 }
1134 return NULL;
1135}
1136
1137/**
1138 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1139 * @wo: wait options
1140 * @ptrace: is the wait for ptrace
1141 * @p: task to wait for
1142 *
1143 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1144 *
1145 * CONTEXT:
1146 * read_lock(&tasklist_lock), which is released if return value is
1147 * non-zero. Also, grabs and releases @p->sighand->siglock.
1148 *
1149 * RETURNS:
1150 * 0 if wait condition didn't exist and search for other wait conditions
1151 * should continue. Non-zero return, -errno on failure and @p's pid on
1152 * success, implies that tasklist_lock is released and wait condition
1153 * search should terminate.
1154 */
1155static int wait_task_stopped(struct wait_opts *wo,
1156 int ptrace, struct task_struct *p)
1157{
1158 struct waitid_info *infop;
1159 int exit_code, *p_code, why;
1160 uid_t uid = 0; /* unneeded, required by compiler */
1161 pid_t pid;
1162
1163 /*
1164 * Traditionally we see ptrace'd stopped tasks regardless of options.
1165 */
1166 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1167 return 0;
1168
1169 if (!task_stopped_code(p, ptrace))
1170 return 0;
1171
1172 exit_code = 0;
1173 spin_lock_irq(&p->sighand->siglock);
1174
1175 p_code = task_stopped_code(p, ptrace);
1176 if (unlikely(!p_code))
1177 goto unlock_sig;
1178
1179 exit_code = *p_code;
1180 if (!exit_code)
1181 goto unlock_sig;
1182
1183 if (!unlikely(wo->wo_flags & WNOWAIT))
1184 *p_code = 0;
1185
1186 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1187unlock_sig:
1188 spin_unlock_irq(&p->sighand->siglock);
1189 if (!exit_code)
1190 return 0;
1191
1192 /*
1193 * Now we are pretty sure this task is interesting.
1194 * Make sure it doesn't get reaped out from under us while we
1195 * give up the lock and then examine it below. We don't want to
1196 * keep holding onto the tasklist_lock while we call getrusage and
1197 * possibly take page faults for user memory.
1198 */
1199 get_task_struct(p);
1200 pid = task_pid_vnr(p);
1201 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1202 read_unlock(&tasklist_lock);
1203 sched_annotate_sleep();
1204 if (wo->wo_rusage)
1205 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1206 put_task_struct(p);
1207
1208 if (likely(!(wo->wo_flags & WNOWAIT)))
1209 wo->wo_stat = (exit_code << 8) | 0x7f;
1210
1211 infop = wo->wo_info;
1212 if (infop) {
1213 infop->cause = why;
1214 infop->status = exit_code;
1215 infop->pid = pid;
1216 infop->uid = uid;
1217 }
1218 return pid;
1219}
1220
1221/*
1222 * Handle do_wait work for one task in a live, non-stopped state.
1223 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1224 * the lock and this task is uninteresting. If we return nonzero, we have
1225 * released the lock and the system call should return.
1226 */
1227static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1228{
1229 struct waitid_info *infop;
1230 pid_t pid;
1231 uid_t uid;
1232
1233 if (!unlikely(wo->wo_flags & WCONTINUED))
1234 return 0;
1235
1236 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1237 return 0;
1238
1239 spin_lock_irq(&p->sighand->siglock);
1240 /* Re-check with the lock held. */
1241 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1242 spin_unlock_irq(&p->sighand->siglock);
1243 return 0;
1244 }
1245 if (!unlikely(wo->wo_flags & WNOWAIT))
1246 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1247 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1248 spin_unlock_irq(&p->sighand->siglock);
1249
1250 pid = task_pid_vnr(p);
1251 get_task_struct(p);
1252 read_unlock(&tasklist_lock);
1253 sched_annotate_sleep();
1254 if (wo->wo_rusage)
1255 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1256 put_task_struct(p);
1257
1258 infop = wo->wo_info;
1259 if (!infop) {
1260 wo->wo_stat = 0xffff;
1261 } else {
1262 infop->cause = CLD_CONTINUED;
1263 infop->pid = pid;
1264 infop->uid = uid;
1265 infop->status = SIGCONT;
1266 }
1267 return pid;
1268}
1269
1270/*
1271 * Consider @p for a wait by @parent.
1272 *
1273 * -ECHILD should be in ->notask_error before the first call.
1274 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1275 * Returns zero if the search for a child should continue;
1276 * then ->notask_error is 0 if @p is an eligible child,
1277 * or still -ECHILD.
1278 */
1279static int wait_consider_task(struct wait_opts *wo, int ptrace,
1280 struct task_struct *p)
1281{
1282 /*
1283 * We can race with wait_task_zombie() from another thread.
1284 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1285 * can't confuse the checks below.
1286 */
1287 int exit_state = READ_ONCE(p->exit_state);
1288 int ret;
1289
1290 if (unlikely(exit_state == EXIT_DEAD))
1291 return 0;
1292
1293 ret = eligible_child(wo, ptrace, p);
1294 if (!ret)
1295 return ret;
1296
1297 if (unlikely(exit_state == EXIT_TRACE)) {
1298 /*
1299 * ptrace == 0 means we are the natural parent. In this case
1300 * we should clear notask_error, debugger will notify us.
1301 */
1302 if (likely(!ptrace))
1303 wo->notask_error = 0;
1304 return 0;
1305 }
1306
1307 if (likely(!ptrace) && unlikely(p->ptrace)) {
1308 /*
1309 * If it is traced by its real parent's group, just pretend
1310 * the caller is ptrace_do_wait() and reap this child if it
1311 * is zombie.
1312 *
1313 * This also hides group stop state from real parent; otherwise
1314 * a single stop can be reported twice as group and ptrace stop.
1315 * If a ptracer wants to distinguish these two events for its
1316 * own children it should create a separate process which takes
1317 * the role of real parent.
1318 */
1319 if (!ptrace_reparented(p))
1320 ptrace = 1;
1321 }
1322
1323 /* slay zombie? */
1324 if (exit_state == EXIT_ZOMBIE) {
1325 /* we don't reap group leaders with subthreads */
1326 if (!delay_group_leader(p)) {
1327 /*
1328 * A zombie ptracee is only visible to its ptracer.
1329 * Notification and reaping will be cascaded to the
1330 * real parent when the ptracer detaches.
1331 */
1332 if (unlikely(ptrace) || likely(!p->ptrace))
1333 return wait_task_zombie(wo, p);
1334 }
1335
1336 /*
1337 * Allow access to stopped/continued state via zombie by
1338 * falling through. Clearing of notask_error is complex.
1339 *
1340 * When !@ptrace:
1341 *
1342 * If WEXITED is set, notask_error should naturally be
1343 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1344 * so, if there are live subthreads, there are events to
1345 * wait for. If all subthreads are dead, it's still safe
1346 * to clear - this function will be called again in finite
1347 * amount time once all the subthreads are released and
1348 * will then return without clearing.
1349 *
1350 * When @ptrace:
1351 *
1352 * Stopped state is per-task and thus can't change once the
1353 * target task dies. Only continued and exited can happen.
1354 * Clear notask_error if WCONTINUED | WEXITED.
1355 */
1356 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1357 wo->notask_error = 0;
1358 } else {
1359 /*
1360 * @p is alive and it's gonna stop, continue or exit, so
1361 * there always is something to wait for.
1362 */
1363 wo->notask_error = 0;
1364 }
1365
1366 /*
1367 * Wait for stopped. Depending on @ptrace, different stopped state
1368 * is used and the two don't interact with each other.
1369 */
1370 ret = wait_task_stopped(wo, ptrace, p);
1371 if (ret)
1372 return ret;
1373
1374 /*
1375 * Wait for continued. There's only one continued state and the
1376 * ptracer can consume it which can confuse the real parent. Don't
1377 * use WCONTINUED from ptracer. You don't need or want it.
1378 */
1379 return wait_task_continued(wo, p);
1380}
1381
1382/*
1383 * Do the work of do_wait() for one thread in the group, @tsk.
1384 *
1385 * -ECHILD should be in ->notask_error before the first call.
1386 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1387 * Returns zero if the search for a child should continue; then
1388 * ->notask_error is 0 if there were any eligible children,
1389 * or still -ECHILD.
1390 */
1391static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1392{
1393 struct task_struct *p;
1394
1395 list_for_each_entry(p, &tsk->children, sibling) {
1396 int ret = wait_consider_task(wo, 0, p);
1397
1398 if (ret)
1399 return ret;
1400 }
1401
1402 return 0;
1403}
1404
1405static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1406{
1407 struct task_struct *p;
1408
1409 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1410 int ret = wait_consider_task(wo, 1, p);
1411
1412 if (ret)
1413 return ret;
1414 }
1415
1416 return 0;
1417}
1418
1419static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1420 int sync, void *key)
1421{
1422 struct wait_opts *wo = container_of(wait, struct wait_opts,
1423 child_wait);
1424 struct task_struct *p = key;
1425
1426 if (!eligible_pid(wo, p))
1427 return 0;
1428
1429 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1430 return 0;
1431
1432 return default_wake_function(wait, mode, sync, key);
1433}
1434
1435void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1436{
1437 __wake_up_sync_key(&parent->signal->wait_chldexit,
1438 TASK_INTERRUPTIBLE, 1, p);
1439}
1440
1441static long do_wait(struct wait_opts *wo)
1442{
1443 struct task_struct *tsk;
1444 int retval;
1445
1446 trace_sched_process_wait(wo->wo_pid);
1447
1448 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1449 wo->child_wait.private = current;
1450 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1451repeat:
1452 /*
1453 * If there is nothing that can match our criteria, just get out.
1454 * We will clear ->notask_error to zero if we see any child that
1455 * might later match our criteria, even if we are not able to reap
1456 * it yet.
1457 */
1458 wo->notask_error = -ECHILD;
1459 if ((wo->wo_type < PIDTYPE_MAX) &&
1460 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1461 goto notask;
1462
1463 set_current_state(TASK_INTERRUPTIBLE);
1464 read_lock(&tasklist_lock);
1465 tsk = current;
1466 do {
1467 retval = do_wait_thread(wo, tsk);
1468 if (retval)
1469 goto end;
1470
1471 retval = ptrace_do_wait(wo, tsk);
1472 if (retval)
1473 goto end;
1474
1475 if (wo->wo_flags & __WNOTHREAD)
1476 break;
1477 } while_each_thread(current, tsk);
1478 read_unlock(&tasklist_lock);
1479
1480notask:
1481 retval = wo->notask_error;
1482 if (!retval && !(wo->wo_flags & WNOHANG)) {
1483 retval = -ERESTARTSYS;
1484 if (!signal_pending(current)) {
1485 schedule();
1486 goto repeat;
1487 }
1488 }
1489end:
1490 __set_current_state(TASK_RUNNING);
1491 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1492 return retval;
1493}
1494
1495static struct pid *pidfd_get_pid(unsigned int fd)
1496{
1497 struct fd f;
1498 struct pid *pid;
1499
1500 f = fdget(fd);
1501 if (!f.file)
1502 return ERR_PTR(-EBADF);
1503
1504 pid = pidfd_pid(f.file);
1505 if (!IS_ERR(pid))
1506 get_pid(pid);
1507
1508 fdput(f);
1509 return pid;
1510}
1511
1512static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1513 int options, struct rusage *ru)
1514{
1515 struct wait_opts wo;
1516 struct pid *pid = NULL;
1517 enum pid_type type;
1518 long ret;
1519
1520 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1521 __WNOTHREAD|__WCLONE|__WALL))
1522 return -EINVAL;
1523 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1524 return -EINVAL;
1525
1526 switch (which) {
1527 case P_ALL:
1528 type = PIDTYPE_MAX;
1529 break;
1530 case P_PID:
1531 type = PIDTYPE_PID;
1532 if (upid <= 0)
1533 return -EINVAL;
1534
1535 pid = find_get_pid(upid);
1536 break;
1537 case P_PGID:
1538 type = PIDTYPE_PGID;
1539 if (upid < 0)
1540 return -EINVAL;
1541
1542 if (upid)
1543 pid = find_get_pid(upid);
1544 else
1545 pid = get_task_pid(current, PIDTYPE_PGID);
1546 break;
1547 case P_PIDFD:
1548 type = PIDTYPE_PID;
1549 if (upid < 0)
1550 return -EINVAL;
1551
1552 pid = pidfd_get_pid(upid);
1553 if (IS_ERR(pid))
1554 return PTR_ERR(pid);
1555 break;
1556 default:
1557 return -EINVAL;
1558 }
1559
1560 wo.wo_type = type;
1561 wo.wo_pid = pid;
1562 wo.wo_flags = options;
1563 wo.wo_info = infop;
1564 wo.wo_rusage = ru;
1565 ret = do_wait(&wo);
1566
1567 put_pid(pid);
1568 return ret;
1569}
1570
1571SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1572 infop, int, options, struct rusage __user *, ru)
1573{
1574 struct rusage r;
1575 struct waitid_info info = {.status = 0};
1576 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1577 int signo = 0;
1578
1579 if (err > 0) {
1580 signo = SIGCHLD;
1581 err = 0;
1582 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1583 return -EFAULT;
1584 }
1585 if (!infop)
1586 return err;
1587
1588 if (!user_access_begin(infop, sizeof(*infop)))
1589 return -EFAULT;
1590
1591 unsafe_put_user(signo, &infop->si_signo, Efault);
1592 unsafe_put_user(0, &infop->si_errno, Efault);
1593 unsafe_put_user(info.cause, &infop->si_code, Efault);
1594 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1595 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1596 unsafe_put_user(info.status, &infop->si_status, Efault);
1597 user_access_end();
1598 return err;
1599Efault:
1600 user_access_end();
1601 return -EFAULT;
1602}
1603
1604long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1605 struct rusage *ru)
1606{
1607 struct wait_opts wo;
1608 struct pid *pid = NULL;
1609 enum pid_type type;
1610 long ret;
1611
1612 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1613 __WNOTHREAD|__WCLONE|__WALL))
1614 return -EINVAL;
1615
1616 /* -INT_MIN is not defined */
1617 if (upid == INT_MIN)
1618 return -ESRCH;
1619
1620 if (upid == -1)
1621 type = PIDTYPE_MAX;
1622 else if (upid < 0) {
1623 type = PIDTYPE_PGID;
1624 pid = find_get_pid(-upid);
1625 } else if (upid == 0) {
1626 type = PIDTYPE_PGID;
1627 pid = get_task_pid(current, PIDTYPE_PGID);
1628 } else /* upid > 0 */ {
1629 type = PIDTYPE_PID;
1630 pid = find_get_pid(upid);
1631 }
1632
1633 wo.wo_type = type;
1634 wo.wo_pid = pid;
1635 wo.wo_flags = options | WEXITED;
1636 wo.wo_info = NULL;
1637 wo.wo_stat = 0;
1638 wo.wo_rusage = ru;
1639 ret = do_wait(&wo);
1640 put_pid(pid);
1641 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1642 ret = -EFAULT;
1643
1644 return ret;
1645}
1646
1647SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1648 int, options, struct rusage __user *, ru)
1649{
1650 struct rusage r;
1651 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1652
1653 if (err > 0) {
1654 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1655 return -EFAULT;
1656 }
1657 return err;
1658}
1659
1660#ifdef __ARCH_WANT_SYS_WAITPID
1661
1662/*
1663 * sys_waitpid() remains for compatibility. waitpid() should be
1664 * implemented by calling sys_wait4() from libc.a.
1665 */
1666SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1667{
1668 return kernel_wait4(pid, stat_addr, options, NULL);
1669}
1670
1671#endif
1672
1673#ifdef CONFIG_COMPAT
1674COMPAT_SYSCALL_DEFINE4(wait4,
1675 compat_pid_t, pid,
1676 compat_uint_t __user *, stat_addr,
1677 int, options,
1678 struct compat_rusage __user *, ru)
1679{
1680 struct rusage r;
1681 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1682 if (err > 0) {
1683 if (ru && put_compat_rusage(&r, ru))
1684 return -EFAULT;
1685 }
1686 return err;
1687}
1688
1689COMPAT_SYSCALL_DEFINE5(waitid,
1690 int, which, compat_pid_t, pid,
1691 struct compat_siginfo __user *, infop, int, options,
1692 struct compat_rusage __user *, uru)
1693{
1694 struct rusage ru;
1695 struct waitid_info info = {.status = 0};
1696 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1697 int signo = 0;
1698 if (err > 0) {
1699 signo = SIGCHLD;
1700 err = 0;
1701 if (uru) {
1702 /* kernel_waitid() overwrites everything in ru */
1703 if (COMPAT_USE_64BIT_TIME)
1704 err = copy_to_user(uru, &ru, sizeof(ru));
1705 else
1706 err = put_compat_rusage(&ru, uru);
1707 if (err)
1708 return -EFAULT;
1709 }
1710 }
1711
1712 if (!infop)
1713 return err;
1714
1715 if (!user_access_begin(infop, sizeof(*infop)))
1716 return -EFAULT;
1717
1718 unsafe_put_user(signo, &infop->si_signo, Efault);
1719 unsafe_put_user(0, &infop->si_errno, Efault);
1720 unsafe_put_user(info.cause, &infop->si_code, Efault);
1721 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1722 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1723 unsafe_put_user(info.status, &infop->si_status, Efault);
1724 user_access_end();
1725 return err;
1726Efault:
1727 user_access_end();
1728 return -EFAULT;
1729}
1730#endif
1731
1732__weak void abort(void)
1733{
1734 BUG();
1735
1736 /* if that doesn't kill us, halt */
1737 panic("Oops failed to kill thread");
1738}
1739EXPORT_SYMBOL(abort);