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/task_io_accounting_ops.h>
52#include <linux/blkdev.h>
53#include <linux/task_work.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/kmsan.h>
64#include <linux/random.h>
65#include <linux/rcuwait.h>
66#include <linux/compat.h>
67#include <linux/io_uring.h>
68#include <linux/kprobes.h>
69#include <linux/rethook.h>
70#include <linux/sysfs.h>
71#include <linux/user_events.h>
72#include <linux/uaccess.h>
73
74#include <uapi/linux/wait.h>
75
76#include <asm/unistd.h>
77#include <asm/mmu_context.h>
78
79#include "exit.h"
80
81/*
82 * The default value should be high enough to not crash a system that randomly
83 * crashes its kernel from time to time, but low enough to at least not permit
84 * overflowing 32-bit refcounts or the ldsem writer count.
85 */
86static unsigned int oops_limit = 10000;
87
88#ifdef CONFIG_SYSCTL
89static struct ctl_table kern_exit_table[] = {
90 {
91 .procname = "oops_limit",
92 .data = &oops_limit,
93 .maxlen = sizeof(oops_limit),
94 .mode = 0644,
95 .proc_handler = proc_douintvec,
96 },
97 { }
98};
99
100static __init int kernel_exit_sysctls_init(void)
101{
102 register_sysctl_init("kernel", kern_exit_table);
103 return 0;
104}
105late_initcall(kernel_exit_sysctls_init);
106#endif
107
108static atomic_t oops_count = ATOMIC_INIT(0);
109
110#ifdef CONFIG_SYSFS
111static ssize_t oops_count_show(struct kobject *kobj, struct kobj_attribute *attr,
112 char *page)
113{
114 return sysfs_emit(page, "%d\n", atomic_read(&oops_count));
115}
116
117static struct kobj_attribute oops_count_attr = __ATTR_RO(oops_count);
118
119static __init int kernel_exit_sysfs_init(void)
120{
121 sysfs_add_file_to_group(kernel_kobj, &oops_count_attr.attr, NULL);
122 return 0;
123}
124late_initcall(kernel_exit_sysfs_init);
125#endif
126
127static void __unhash_process(struct task_struct *p, bool group_dead)
128{
129 nr_threads--;
130 detach_pid(p, PIDTYPE_PID);
131 if (group_dead) {
132 detach_pid(p, PIDTYPE_TGID);
133 detach_pid(p, PIDTYPE_PGID);
134 detach_pid(p, PIDTYPE_SID);
135
136 list_del_rcu(&p->tasks);
137 list_del_init(&p->sibling);
138 __this_cpu_dec(process_counts);
139 }
140 list_del_rcu(&p->thread_node);
141}
142
143/*
144 * This function expects the tasklist_lock write-locked.
145 */
146static void __exit_signal(struct task_struct *tsk)
147{
148 struct signal_struct *sig = tsk->signal;
149 bool group_dead = thread_group_leader(tsk);
150 struct sighand_struct *sighand;
151 struct tty_struct *tty;
152 u64 utime, stime;
153
154 sighand = rcu_dereference_check(tsk->sighand,
155 lockdep_tasklist_lock_is_held());
156 spin_lock(&sighand->siglock);
157
158#ifdef CONFIG_POSIX_TIMERS
159 posix_cpu_timers_exit(tsk);
160 if (group_dead)
161 posix_cpu_timers_exit_group(tsk);
162#endif
163
164 if (group_dead) {
165 tty = sig->tty;
166 sig->tty = NULL;
167 } else {
168 /*
169 * If there is any task waiting for the group exit
170 * then notify it:
171 */
172 if (sig->notify_count > 0 && !--sig->notify_count)
173 wake_up_process(sig->group_exec_task);
174
175 if (tsk == sig->curr_target)
176 sig->curr_target = next_thread(tsk);
177 }
178
179 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
180 sizeof(unsigned long long));
181
182 /*
183 * Accumulate here the counters for all threads as they die. We could
184 * skip the group leader because it is the last user of signal_struct,
185 * but we want to avoid the race with thread_group_cputime() which can
186 * see the empty ->thread_head list.
187 */
188 task_cputime(tsk, &utime, &stime);
189 write_seqlock(&sig->stats_lock);
190 sig->utime += utime;
191 sig->stime += stime;
192 sig->gtime += task_gtime(tsk);
193 sig->min_flt += tsk->min_flt;
194 sig->maj_flt += tsk->maj_flt;
195 sig->nvcsw += tsk->nvcsw;
196 sig->nivcsw += tsk->nivcsw;
197 sig->inblock += task_io_get_inblock(tsk);
198 sig->oublock += task_io_get_oublock(tsk);
199 task_io_accounting_add(&sig->ioac, &tsk->ioac);
200 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
201 sig->nr_threads--;
202 __unhash_process(tsk, group_dead);
203 write_sequnlock(&sig->stats_lock);
204
205 /*
206 * Do this under ->siglock, we can race with another thread
207 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
208 */
209 flush_sigqueue(&tsk->pending);
210 tsk->sighand = NULL;
211 spin_unlock(&sighand->siglock);
212
213 __cleanup_sighand(sighand);
214 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
215 if (group_dead) {
216 flush_sigqueue(&sig->shared_pending);
217 tty_kref_put(tty);
218 }
219}
220
221static void delayed_put_task_struct(struct rcu_head *rhp)
222{
223 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
224
225 kprobe_flush_task(tsk);
226 rethook_flush_task(tsk);
227 perf_event_delayed_put(tsk);
228 trace_sched_process_free(tsk);
229 put_task_struct(tsk);
230}
231
232void put_task_struct_rcu_user(struct task_struct *task)
233{
234 if (refcount_dec_and_test(&task->rcu_users))
235 call_rcu(&task->rcu, delayed_put_task_struct);
236}
237
238void __weak release_thread(struct task_struct *dead_task)
239{
240}
241
242void release_task(struct task_struct *p)
243{
244 struct task_struct *leader;
245 struct pid *thread_pid;
246 int zap_leader;
247repeat:
248 /* don't need to get the RCU readlock here - the process is dead and
249 * can't be modifying its own credentials. But shut RCU-lockdep up */
250 rcu_read_lock();
251 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
252 rcu_read_unlock();
253
254 cgroup_release(p);
255
256 write_lock_irq(&tasklist_lock);
257 ptrace_release_task(p);
258 thread_pid = get_pid(p->thread_pid);
259 __exit_signal(p);
260
261 /*
262 * If we are the last non-leader member of the thread
263 * group, and the leader is zombie, then notify the
264 * group leader's parent process. (if it wants notification.)
265 */
266 zap_leader = 0;
267 leader = p->group_leader;
268 if (leader != p && thread_group_empty(leader)
269 && leader->exit_state == EXIT_ZOMBIE) {
270 /*
271 * If we were the last child thread and the leader has
272 * exited already, and the leader's parent ignores SIGCHLD,
273 * then we are the one who should release the leader.
274 */
275 zap_leader = do_notify_parent(leader, leader->exit_signal);
276 if (zap_leader)
277 leader->exit_state = EXIT_DEAD;
278 }
279
280 write_unlock_irq(&tasklist_lock);
281 seccomp_filter_release(p);
282 proc_flush_pid(thread_pid);
283 put_pid(thread_pid);
284 release_thread(p);
285 put_task_struct_rcu_user(p);
286
287 p = leader;
288 if (unlikely(zap_leader))
289 goto repeat;
290}
291
292int rcuwait_wake_up(struct rcuwait *w)
293{
294 int ret = 0;
295 struct task_struct *task;
296
297 rcu_read_lock();
298
299 /*
300 * Order condition vs @task, such that everything prior to the load
301 * of @task is visible. This is the condition as to why the user called
302 * rcuwait_wake() in the first place. Pairs with set_current_state()
303 * barrier (A) in rcuwait_wait_event().
304 *
305 * WAIT WAKE
306 * [S] tsk = current [S] cond = true
307 * MB (A) MB (B)
308 * [L] cond [L] tsk
309 */
310 smp_mb(); /* (B) */
311
312 task = rcu_dereference(w->task);
313 if (task)
314 ret = wake_up_process(task);
315 rcu_read_unlock();
316
317 return ret;
318}
319EXPORT_SYMBOL_GPL(rcuwait_wake_up);
320
321/*
322 * Determine if a process group is "orphaned", according to the POSIX
323 * definition in 2.2.2.52. Orphaned process groups are not to be affected
324 * by terminal-generated stop signals. Newly orphaned process groups are
325 * to receive a SIGHUP and a SIGCONT.
326 *
327 * "I ask you, have you ever known what it is to be an orphan?"
328 */
329static int will_become_orphaned_pgrp(struct pid *pgrp,
330 struct task_struct *ignored_task)
331{
332 struct task_struct *p;
333
334 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
335 if ((p == ignored_task) ||
336 (p->exit_state && thread_group_empty(p)) ||
337 is_global_init(p->real_parent))
338 continue;
339
340 if (task_pgrp(p->real_parent) != pgrp &&
341 task_session(p->real_parent) == task_session(p))
342 return 0;
343 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
344
345 return 1;
346}
347
348int is_current_pgrp_orphaned(void)
349{
350 int retval;
351
352 read_lock(&tasklist_lock);
353 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
354 read_unlock(&tasklist_lock);
355
356 return retval;
357}
358
359static bool has_stopped_jobs(struct pid *pgrp)
360{
361 struct task_struct *p;
362
363 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
364 if (p->signal->flags & SIGNAL_STOP_STOPPED)
365 return true;
366 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
367
368 return false;
369}
370
371/*
372 * Check to see if any process groups have become orphaned as
373 * a result of our exiting, and if they have any stopped jobs,
374 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
375 */
376static void
377kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
378{
379 struct pid *pgrp = task_pgrp(tsk);
380 struct task_struct *ignored_task = tsk;
381
382 if (!parent)
383 /* exit: our father is in a different pgrp than
384 * we are and we were the only connection outside.
385 */
386 parent = tsk->real_parent;
387 else
388 /* reparent: our child is in a different pgrp than
389 * we are, and it was the only connection outside.
390 */
391 ignored_task = NULL;
392
393 if (task_pgrp(parent) != pgrp &&
394 task_session(parent) == task_session(tsk) &&
395 will_become_orphaned_pgrp(pgrp, ignored_task) &&
396 has_stopped_jobs(pgrp)) {
397 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
398 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
399 }
400}
401
402static void coredump_task_exit(struct task_struct *tsk)
403{
404 struct core_state *core_state;
405
406 /*
407 * Serialize with any possible pending coredump.
408 * We must hold siglock around checking core_state
409 * and setting PF_POSTCOREDUMP. The core-inducing thread
410 * will increment ->nr_threads for each thread in the
411 * group without PF_POSTCOREDUMP set.
412 */
413 spin_lock_irq(&tsk->sighand->siglock);
414 tsk->flags |= PF_POSTCOREDUMP;
415 core_state = tsk->signal->core_state;
416 spin_unlock_irq(&tsk->sighand->siglock);
417
418 /* The vhost_worker does not particpate in coredumps */
419 if (core_state &&
420 ((tsk->flags & (PF_IO_WORKER | PF_USER_WORKER)) != PF_USER_WORKER)) {
421 struct core_thread self;
422
423 self.task = current;
424 if (self.task->flags & PF_SIGNALED)
425 self.next = xchg(&core_state->dumper.next, &self);
426 else
427 self.task = NULL;
428 /*
429 * Implies mb(), the result of xchg() must be visible
430 * to core_state->dumper.
431 */
432 if (atomic_dec_and_test(&core_state->nr_threads))
433 complete(&core_state->startup);
434
435 for (;;) {
436 set_current_state(TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
437 if (!self.task) /* see coredump_finish() */
438 break;
439 schedule();
440 }
441 __set_current_state(TASK_RUNNING);
442 }
443}
444
445#ifdef CONFIG_MEMCG
446/*
447 * A task is exiting. If it owned this mm, find a new owner for the mm.
448 */
449void mm_update_next_owner(struct mm_struct *mm)
450{
451 struct task_struct *c, *g, *p = current;
452
453retry:
454 /*
455 * If the exiting or execing task is not the owner, it's
456 * someone else's problem.
457 */
458 if (mm->owner != p)
459 return;
460 /*
461 * The current owner is exiting/execing and there are no other
462 * candidates. Do not leave the mm pointing to a possibly
463 * freed task structure.
464 */
465 if (atomic_read(&mm->mm_users) <= 1) {
466 WRITE_ONCE(mm->owner, NULL);
467 return;
468 }
469
470 read_lock(&tasklist_lock);
471 /*
472 * Search in the children
473 */
474 list_for_each_entry(c, &p->children, sibling) {
475 if (c->mm == mm)
476 goto assign_new_owner;
477 }
478
479 /*
480 * Search in the siblings
481 */
482 list_for_each_entry(c, &p->real_parent->children, sibling) {
483 if (c->mm == mm)
484 goto assign_new_owner;
485 }
486
487 /*
488 * Search through everything else, we should not get here often.
489 */
490 for_each_process(g) {
491 if (g->flags & PF_KTHREAD)
492 continue;
493 for_each_thread(g, c) {
494 if (c->mm == mm)
495 goto assign_new_owner;
496 if (c->mm)
497 break;
498 }
499 }
500 read_unlock(&tasklist_lock);
501 /*
502 * We found no owner yet mm_users > 1: this implies that we are
503 * most likely racing with swapoff (try_to_unuse()) or /proc or
504 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
505 */
506 WRITE_ONCE(mm->owner, NULL);
507 return;
508
509assign_new_owner:
510 BUG_ON(c == p);
511 get_task_struct(c);
512 /*
513 * The task_lock protects c->mm from changing.
514 * We always want mm->owner->mm == mm
515 */
516 task_lock(c);
517 /*
518 * Delay read_unlock() till we have the task_lock()
519 * to ensure that c does not slip away underneath us
520 */
521 read_unlock(&tasklist_lock);
522 if (c->mm != mm) {
523 task_unlock(c);
524 put_task_struct(c);
525 goto retry;
526 }
527 WRITE_ONCE(mm->owner, c);
528 lru_gen_migrate_mm(mm);
529 task_unlock(c);
530 put_task_struct(c);
531}
532#endif /* CONFIG_MEMCG */
533
534/*
535 * Turn us into a lazy TLB process if we
536 * aren't already..
537 */
538static void exit_mm(void)
539{
540 struct mm_struct *mm = current->mm;
541
542 exit_mm_release(current, mm);
543 if (!mm)
544 return;
545 mmap_read_lock(mm);
546 mmgrab_lazy_tlb(mm);
547 BUG_ON(mm != current->active_mm);
548 /* more a memory barrier than a real lock */
549 task_lock(current);
550 /*
551 * When a thread stops operating on an address space, the loop
552 * in membarrier_private_expedited() may not observe that
553 * tsk->mm, and the loop in membarrier_global_expedited() may
554 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
555 * rq->membarrier_state, so those would not issue an IPI.
556 * Membarrier requires a memory barrier after accessing
557 * user-space memory, before clearing tsk->mm or the
558 * rq->membarrier_state.
559 */
560 smp_mb__after_spinlock();
561 local_irq_disable();
562 current->mm = NULL;
563 membarrier_update_current_mm(NULL);
564 enter_lazy_tlb(mm, current);
565 local_irq_enable();
566 task_unlock(current);
567 mmap_read_unlock(mm);
568 mm_update_next_owner(mm);
569 mmput(mm);
570 if (test_thread_flag(TIF_MEMDIE))
571 exit_oom_victim();
572}
573
574static struct task_struct *find_alive_thread(struct task_struct *p)
575{
576 struct task_struct *t;
577
578 for_each_thread(p, t) {
579 if (!(t->flags & PF_EXITING))
580 return t;
581 }
582 return NULL;
583}
584
585static struct task_struct *find_child_reaper(struct task_struct *father,
586 struct list_head *dead)
587 __releases(&tasklist_lock)
588 __acquires(&tasklist_lock)
589{
590 struct pid_namespace *pid_ns = task_active_pid_ns(father);
591 struct task_struct *reaper = pid_ns->child_reaper;
592 struct task_struct *p, *n;
593
594 if (likely(reaper != father))
595 return reaper;
596
597 reaper = find_alive_thread(father);
598 if (reaper) {
599 pid_ns->child_reaper = reaper;
600 return reaper;
601 }
602
603 write_unlock_irq(&tasklist_lock);
604
605 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
606 list_del_init(&p->ptrace_entry);
607 release_task(p);
608 }
609
610 zap_pid_ns_processes(pid_ns);
611 write_lock_irq(&tasklist_lock);
612
613 return father;
614}
615
616/*
617 * When we die, we re-parent all our children, and try to:
618 * 1. give them to another thread in our thread group, if such a member exists
619 * 2. give it to the first ancestor process which prctl'd itself as a
620 * child_subreaper for its children (like a service manager)
621 * 3. give it to the init process (PID 1) in our pid namespace
622 */
623static struct task_struct *find_new_reaper(struct task_struct *father,
624 struct task_struct *child_reaper)
625{
626 struct task_struct *thread, *reaper;
627
628 thread = find_alive_thread(father);
629 if (thread)
630 return thread;
631
632 if (father->signal->has_child_subreaper) {
633 unsigned int ns_level = task_pid(father)->level;
634 /*
635 * Find the first ->is_child_subreaper ancestor in our pid_ns.
636 * We can't check reaper != child_reaper to ensure we do not
637 * cross the namespaces, the exiting parent could be injected
638 * by setns() + fork().
639 * We check pid->level, this is slightly more efficient than
640 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
641 */
642 for (reaper = father->real_parent;
643 task_pid(reaper)->level == ns_level;
644 reaper = reaper->real_parent) {
645 if (reaper == &init_task)
646 break;
647 if (!reaper->signal->is_child_subreaper)
648 continue;
649 thread = find_alive_thread(reaper);
650 if (thread)
651 return thread;
652 }
653 }
654
655 return child_reaper;
656}
657
658/*
659* Any that need to be release_task'd are put on the @dead list.
660 */
661static void reparent_leader(struct task_struct *father, struct task_struct *p,
662 struct list_head *dead)
663{
664 if (unlikely(p->exit_state == EXIT_DEAD))
665 return;
666
667 /* We don't want people slaying init. */
668 p->exit_signal = SIGCHLD;
669
670 /* If it has exited notify the new parent about this child's death. */
671 if (!p->ptrace &&
672 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
673 if (do_notify_parent(p, p->exit_signal)) {
674 p->exit_state = EXIT_DEAD;
675 list_add(&p->ptrace_entry, dead);
676 }
677 }
678
679 kill_orphaned_pgrp(p, father);
680}
681
682/*
683 * This does two things:
684 *
685 * A. Make init inherit all the child processes
686 * B. Check to see if any process groups have become orphaned
687 * as a result of our exiting, and if they have any stopped
688 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
689 */
690static void forget_original_parent(struct task_struct *father,
691 struct list_head *dead)
692{
693 struct task_struct *p, *t, *reaper;
694
695 if (unlikely(!list_empty(&father->ptraced)))
696 exit_ptrace(father, dead);
697
698 /* Can drop and reacquire tasklist_lock */
699 reaper = find_child_reaper(father, dead);
700 if (list_empty(&father->children))
701 return;
702
703 reaper = find_new_reaper(father, reaper);
704 list_for_each_entry(p, &father->children, sibling) {
705 for_each_thread(p, t) {
706 RCU_INIT_POINTER(t->real_parent, reaper);
707 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
708 if (likely(!t->ptrace))
709 t->parent = t->real_parent;
710 if (t->pdeath_signal)
711 group_send_sig_info(t->pdeath_signal,
712 SEND_SIG_NOINFO, t,
713 PIDTYPE_TGID);
714 }
715 /*
716 * If this is a threaded reparent there is no need to
717 * notify anyone anything has happened.
718 */
719 if (!same_thread_group(reaper, father))
720 reparent_leader(father, p, dead);
721 }
722 list_splice_tail_init(&father->children, &reaper->children);
723}
724
725/*
726 * Send signals to all our closest relatives so that they know
727 * to properly mourn us..
728 */
729static void exit_notify(struct task_struct *tsk, int group_dead)
730{
731 bool autoreap;
732 struct task_struct *p, *n;
733 LIST_HEAD(dead);
734
735 write_lock_irq(&tasklist_lock);
736 forget_original_parent(tsk, &dead);
737
738 if (group_dead)
739 kill_orphaned_pgrp(tsk->group_leader, NULL);
740
741 tsk->exit_state = EXIT_ZOMBIE;
742 /*
743 * sub-thread or delay_group_leader(), wake up the
744 * PIDFD_THREAD waiters.
745 */
746 if (!thread_group_empty(tsk))
747 do_notify_pidfd(tsk);
748
749 if (unlikely(tsk->ptrace)) {
750 int sig = thread_group_leader(tsk) &&
751 thread_group_empty(tsk) &&
752 !ptrace_reparented(tsk) ?
753 tsk->exit_signal : SIGCHLD;
754 autoreap = do_notify_parent(tsk, sig);
755 } else if (thread_group_leader(tsk)) {
756 autoreap = thread_group_empty(tsk) &&
757 do_notify_parent(tsk, tsk->exit_signal);
758 } else {
759 autoreap = true;
760 }
761
762 if (autoreap) {
763 tsk->exit_state = EXIT_DEAD;
764 list_add(&tsk->ptrace_entry, &dead);
765 }
766
767 /* mt-exec, de_thread() is waiting for group leader */
768 if (unlikely(tsk->signal->notify_count < 0))
769 wake_up_process(tsk->signal->group_exec_task);
770 write_unlock_irq(&tasklist_lock);
771
772 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
773 list_del_init(&p->ptrace_entry);
774 release_task(p);
775 }
776}
777
778#ifdef CONFIG_DEBUG_STACK_USAGE
779static void check_stack_usage(void)
780{
781 static DEFINE_SPINLOCK(low_water_lock);
782 static int lowest_to_date = THREAD_SIZE;
783 unsigned long free;
784
785 free = stack_not_used(current);
786
787 if (free >= lowest_to_date)
788 return;
789
790 spin_lock(&low_water_lock);
791 if (free < lowest_to_date) {
792 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
793 current->comm, task_pid_nr(current), free);
794 lowest_to_date = free;
795 }
796 spin_unlock(&low_water_lock);
797}
798#else
799static inline void check_stack_usage(void) {}
800#endif
801
802static void synchronize_group_exit(struct task_struct *tsk, long code)
803{
804 struct sighand_struct *sighand = tsk->sighand;
805 struct signal_struct *signal = tsk->signal;
806
807 spin_lock_irq(&sighand->siglock);
808 signal->quick_threads--;
809 if ((signal->quick_threads == 0) &&
810 !(signal->flags & SIGNAL_GROUP_EXIT)) {
811 signal->flags = SIGNAL_GROUP_EXIT;
812 signal->group_exit_code = code;
813 signal->group_stop_count = 0;
814 }
815 spin_unlock_irq(&sighand->siglock);
816}
817
818void __noreturn do_exit(long code)
819{
820 struct task_struct *tsk = current;
821 int group_dead;
822
823 WARN_ON(irqs_disabled());
824
825 synchronize_group_exit(tsk, code);
826
827 WARN_ON(tsk->plug);
828
829 kcov_task_exit(tsk);
830 kmsan_task_exit(tsk);
831
832 coredump_task_exit(tsk);
833 ptrace_event(PTRACE_EVENT_EXIT, code);
834 user_events_exit(tsk);
835
836 io_uring_files_cancel();
837 exit_signals(tsk); /* sets PF_EXITING */
838
839 acct_update_integrals(tsk);
840 group_dead = atomic_dec_and_test(&tsk->signal->live);
841 if (group_dead) {
842 /*
843 * If the last thread of global init has exited, panic
844 * immediately to get a useable coredump.
845 */
846 if (unlikely(is_global_init(tsk)))
847 panic("Attempted to kill init! exitcode=0x%08x\n",
848 tsk->signal->group_exit_code ?: (int)code);
849
850#ifdef CONFIG_POSIX_TIMERS
851 hrtimer_cancel(&tsk->signal->real_timer);
852 exit_itimers(tsk);
853#endif
854 if (tsk->mm)
855 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
856 }
857 acct_collect(code, group_dead);
858 if (group_dead)
859 tty_audit_exit();
860 audit_free(tsk);
861
862 tsk->exit_code = code;
863 taskstats_exit(tsk, group_dead);
864
865 exit_mm();
866
867 if (group_dead)
868 acct_process();
869 trace_sched_process_exit(tsk);
870
871 exit_sem(tsk);
872 exit_shm(tsk);
873 exit_files(tsk);
874 exit_fs(tsk);
875 if (group_dead)
876 disassociate_ctty(1);
877 exit_task_namespaces(tsk);
878 exit_task_work(tsk);
879 exit_thread(tsk);
880
881 /*
882 * Flush inherited counters to the parent - before the parent
883 * gets woken up by child-exit notifications.
884 *
885 * because of cgroup mode, must be called before cgroup_exit()
886 */
887 perf_event_exit_task(tsk);
888
889 sched_autogroup_exit_task(tsk);
890 cgroup_exit(tsk);
891
892 /*
893 * FIXME: do that only when needed, using sched_exit tracepoint
894 */
895 flush_ptrace_hw_breakpoint(tsk);
896
897 exit_tasks_rcu_start();
898 exit_notify(tsk, group_dead);
899 proc_exit_connector(tsk);
900 mpol_put_task_policy(tsk);
901#ifdef CONFIG_FUTEX
902 if (unlikely(current->pi_state_cache))
903 kfree(current->pi_state_cache);
904#endif
905 /*
906 * Make sure we are holding no locks:
907 */
908 debug_check_no_locks_held();
909
910 if (tsk->io_context)
911 exit_io_context(tsk);
912
913 if (tsk->splice_pipe)
914 free_pipe_info(tsk->splice_pipe);
915
916 if (tsk->task_frag.page)
917 put_page(tsk->task_frag.page);
918
919 exit_task_stack_account(tsk);
920
921 check_stack_usage();
922 preempt_disable();
923 if (tsk->nr_dirtied)
924 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
925 exit_rcu();
926 exit_tasks_rcu_finish();
927
928 lockdep_free_task(tsk);
929 do_task_dead();
930}
931
932void __noreturn make_task_dead(int signr)
933{
934 /*
935 * Take the task off the cpu after something catastrophic has
936 * happened.
937 *
938 * We can get here from a kernel oops, sometimes with preemption off.
939 * Start by checking for critical errors.
940 * Then fix up important state like USER_DS and preemption.
941 * Then do everything else.
942 */
943 struct task_struct *tsk = current;
944 unsigned int limit;
945
946 if (unlikely(in_interrupt()))
947 panic("Aiee, killing interrupt handler!");
948 if (unlikely(!tsk->pid))
949 panic("Attempted to kill the idle task!");
950
951 if (unlikely(irqs_disabled())) {
952 pr_info("note: %s[%d] exited with irqs disabled\n",
953 current->comm, task_pid_nr(current));
954 local_irq_enable();
955 }
956 if (unlikely(in_atomic())) {
957 pr_info("note: %s[%d] exited with preempt_count %d\n",
958 current->comm, task_pid_nr(current),
959 preempt_count());
960 preempt_count_set(PREEMPT_ENABLED);
961 }
962
963 /*
964 * Every time the system oopses, if the oops happens while a reference
965 * to an object was held, the reference leaks.
966 * If the oops doesn't also leak memory, repeated oopsing can cause
967 * reference counters to wrap around (if they're not using refcount_t).
968 * This means that repeated oopsing can make unexploitable-looking bugs
969 * exploitable through repeated oopsing.
970 * To make sure this can't happen, place an upper bound on how often the
971 * kernel may oops without panic().
972 */
973 limit = READ_ONCE(oops_limit);
974 if (atomic_inc_return(&oops_count) >= limit && limit)
975 panic("Oopsed too often (kernel.oops_limit is %d)", limit);
976
977 /*
978 * We're taking recursive faults here in make_task_dead. Safest is to just
979 * leave this task alone and wait for reboot.
980 */
981 if (unlikely(tsk->flags & PF_EXITING)) {
982 pr_alert("Fixing recursive fault but reboot is needed!\n");
983 futex_exit_recursive(tsk);
984 tsk->exit_state = EXIT_DEAD;
985 refcount_inc(&tsk->rcu_users);
986 do_task_dead();
987 }
988
989 do_exit(signr);
990}
991
992SYSCALL_DEFINE1(exit, int, error_code)
993{
994 do_exit((error_code&0xff)<<8);
995}
996
997/*
998 * Take down every thread in the group. This is called by fatal signals
999 * as well as by sys_exit_group (below).
1000 */
1001void __noreturn
1002do_group_exit(int exit_code)
1003{
1004 struct signal_struct *sig = current->signal;
1005
1006 if (sig->flags & SIGNAL_GROUP_EXIT)
1007 exit_code = sig->group_exit_code;
1008 else if (sig->group_exec_task)
1009 exit_code = 0;
1010 else {
1011 struct sighand_struct *const sighand = current->sighand;
1012
1013 spin_lock_irq(&sighand->siglock);
1014 if (sig->flags & SIGNAL_GROUP_EXIT)
1015 /* Another thread got here before we took the lock. */
1016 exit_code = sig->group_exit_code;
1017 else if (sig->group_exec_task)
1018 exit_code = 0;
1019 else {
1020 sig->group_exit_code = exit_code;
1021 sig->flags = SIGNAL_GROUP_EXIT;
1022 zap_other_threads(current);
1023 }
1024 spin_unlock_irq(&sighand->siglock);
1025 }
1026
1027 do_exit(exit_code);
1028 /* NOTREACHED */
1029}
1030
1031/*
1032 * this kills every thread in the thread group. Note that any externally
1033 * wait4()-ing process will get the correct exit code - even if this
1034 * thread is not the thread group leader.
1035 */
1036SYSCALL_DEFINE1(exit_group, int, error_code)
1037{
1038 do_group_exit((error_code & 0xff) << 8);
1039 /* NOTREACHED */
1040 return 0;
1041}
1042
1043static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1044{
1045 return wo->wo_type == PIDTYPE_MAX ||
1046 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1047}
1048
1049static int
1050eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1051{
1052 if (!eligible_pid(wo, p))
1053 return 0;
1054
1055 /*
1056 * Wait for all children (clone and not) if __WALL is set or
1057 * if it is traced by us.
1058 */
1059 if (ptrace || (wo->wo_flags & __WALL))
1060 return 1;
1061
1062 /*
1063 * Otherwise, wait for clone children *only* if __WCLONE is set;
1064 * otherwise, wait for non-clone children *only*.
1065 *
1066 * Note: a "clone" child here is one that reports to its parent
1067 * using a signal other than SIGCHLD, or a non-leader thread which
1068 * we can only see if it is traced by us.
1069 */
1070 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1071 return 0;
1072
1073 return 1;
1074}
1075
1076/*
1077 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1078 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1079 * the lock and this task is uninteresting. If we return nonzero, we have
1080 * released the lock and the system call should return.
1081 */
1082static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1083{
1084 int state, status;
1085 pid_t pid = task_pid_vnr(p);
1086 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1087 struct waitid_info *infop;
1088
1089 if (!likely(wo->wo_flags & WEXITED))
1090 return 0;
1091
1092 if (unlikely(wo->wo_flags & WNOWAIT)) {
1093 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1094 ? p->signal->group_exit_code : p->exit_code;
1095 get_task_struct(p);
1096 read_unlock(&tasklist_lock);
1097 sched_annotate_sleep();
1098 if (wo->wo_rusage)
1099 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1100 put_task_struct(p);
1101 goto out_info;
1102 }
1103 /*
1104 * Move the task's state to DEAD/TRACE, only one thread can do this.
1105 */
1106 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1107 EXIT_TRACE : EXIT_DEAD;
1108 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1109 return 0;
1110 /*
1111 * We own this thread, nobody else can reap it.
1112 */
1113 read_unlock(&tasklist_lock);
1114 sched_annotate_sleep();
1115
1116 /*
1117 * Check thread_group_leader() to exclude the traced sub-threads.
1118 */
1119 if (state == EXIT_DEAD && thread_group_leader(p)) {
1120 struct signal_struct *sig = p->signal;
1121 struct signal_struct *psig = current->signal;
1122 unsigned long maxrss;
1123 u64 tgutime, tgstime;
1124
1125 /*
1126 * The resource counters for the group leader are in its
1127 * own task_struct. Those for dead threads in the group
1128 * are in its signal_struct, as are those for the child
1129 * processes it has previously reaped. All these
1130 * accumulate in the parent's signal_struct c* fields.
1131 *
1132 * We don't bother to take a lock here to protect these
1133 * p->signal fields because the whole thread group is dead
1134 * and nobody can change them.
1135 *
1136 * psig->stats_lock also protects us from our sub-threads
1137 * which can reap other children at the same time.
1138 *
1139 * We use thread_group_cputime_adjusted() to get times for
1140 * the thread group, which consolidates times for all threads
1141 * in the group including the group leader.
1142 */
1143 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1144 write_seqlock_irq(&psig->stats_lock);
1145 psig->cutime += tgutime + sig->cutime;
1146 psig->cstime += tgstime + sig->cstime;
1147 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1148 psig->cmin_flt +=
1149 p->min_flt + sig->min_flt + sig->cmin_flt;
1150 psig->cmaj_flt +=
1151 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1152 psig->cnvcsw +=
1153 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1154 psig->cnivcsw +=
1155 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1156 psig->cinblock +=
1157 task_io_get_inblock(p) +
1158 sig->inblock + sig->cinblock;
1159 psig->coublock +=
1160 task_io_get_oublock(p) +
1161 sig->oublock + sig->coublock;
1162 maxrss = max(sig->maxrss, sig->cmaxrss);
1163 if (psig->cmaxrss < maxrss)
1164 psig->cmaxrss = maxrss;
1165 task_io_accounting_add(&psig->ioac, &p->ioac);
1166 task_io_accounting_add(&psig->ioac, &sig->ioac);
1167 write_sequnlock_irq(&psig->stats_lock);
1168 }
1169
1170 if (wo->wo_rusage)
1171 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1172 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1173 ? p->signal->group_exit_code : p->exit_code;
1174 wo->wo_stat = status;
1175
1176 if (state == EXIT_TRACE) {
1177 write_lock_irq(&tasklist_lock);
1178 /* We dropped tasklist, ptracer could die and untrace */
1179 ptrace_unlink(p);
1180
1181 /* If parent wants a zombie, don't release it now */
1182 state = EXIT_ZOMBIE;
1183 if (do_notify_parent(p, p->exit_signal))
1184 state = EXIT_DEAD;
1185 p->exit_state = state;
1186 write_unlock_irq(&tasklist_lock);
1187 }
1188 if (state == EXIT_DEAD)
1189 release_task(p);
1190
1191out_info:
1192 infop = wo->wo_info;
1193 if (infop) {
1194 if ((status & 0x7f) == 0) {
1195 infop->cause = CLD_EXITED;
1196 infop->status = status >> 8;
1197 } else {
1198 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1199 infop->status = status & 0x7f;
1200 }
1201 infop->pid = pid;
1202 infop->uid = uid;
1203 }
1204
1205 return pid;
1206}
1207
1208static int *task_stopped_code(struct task_struct *p, bool ptrace)
1209{
1210 if (ptrace) {
1211 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1212 return &p->exit_code;
1213 } else {
1214 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1215 return &p->signal->group_exit_code;
1216 }
1217 return NULL;
1218}
1219
1220/**
1221 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1222 * @wo: wait options
1223 * @ptrace: is the wait for ptrace
1224 * @p: task to wait for
1225 *
1226 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1227 *
1228 * CONTEXT:
1229 * read_lock(&tasklist_lock), which is released if return value is
1230 * non-zero. Also, grabs and releases @p->sighand->siglock.
1231 *
1232 * RETURNS:
1233 * 0 if wait condition didn't exist and search for other wait conditions
1234 * should continue. Non-zero return, -errno on failure and @p's pid on
1235 * success, implies that tasklist_lock is released and wait condition
1236 * search should terminate.
1237 */
1238static int wait_task_stopped(struct wait_opts *wo,
1239 int ptrace, struct task_struct *p)
1240{
1241 struct waitid_info *infop;
1242 int exit_code, *p_code, why;
1243 uid_t uid = 0; /* unneeded, required by compiler */
1244 pid_t pid;
1245
1246 /*
1247 * Traditionally we see ptrace'd stopped tasks regardless of options.
1248 */
1249 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1250 return 0;
1251
1252 if (!task_stopped_code(p, ptrace))
1253 return 0;
1254
1255 exit_code = 0;
1256 spin_lock_irq(&p->sighand->siglock);
1257
1258 p_code = task_stopped_code(p, ptrace);
1259 if (unlikely(!p_code))
1260 goto unlock_sig;
1261
1262 exit_code = *p_code;
1263 if (!exit_code)
1264 goto unlock_sig;
1265
1266 if (!unlikely(wo->wo_flags & WNOWAIT))
1267 *p_code = 0;
1268
1269 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1270unlock_sig:
1271 spin_unlock_irq(&p->sighand->siglock);
1272 if (!exit_code)
1273 return 0;
1274
1275 /*
1276 * Now we are pretty sure this task is interesting.
1277 * Make sure it doesn't get reaped out from under us while we
1278 * give up the lock and then examine it below. We don't want to
1279 * keep holding onto the tasklist_lock while we call getrusage and
1280 * possibly take page faults for user memory.
1281 */
1282 get_task_struct(p);
1283 pid = task_pid_vnr(p);
1284 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1285 read_unlock(&tasklist_lock);
1286 sched_annotate_sleep();
1287 if (wo->wo_rusage)
1288 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1289 put_task_struct(p);
1290
1291 if (likely(!(wo->wo_flags & WNOWAIT)))
1292 wo->wo_stat = (exit_code << 8) | 0x7f;
1293
1294 infop = wo->wo_info;
1295 if (infop) {
1296 infop->cause = why;
1297 infop->status = exit_code;
1298 infop->pid = pid;
1299 infop->uid = uid;
1300 }
1301 return pid;
1302}
1303
1304/*
1305 * Handle do_wait work for one task in a live, non-stopped state.
1306 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1307 * the lock and this task is uninteresting. If we return nonzero, we have
1308 * released the lock and the system call should return.
1309 */
1310static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1311{
1312 struct waitid_info *infop;
1313 pid_t pid;
1314 uid_t uid;
1315
1316 if (!unlikely(wo->wo_flags & WCONTINUED))
1317 return 0;
1318
1319 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1320 return 0;
1321
1322 spin_lock_irq(&p->sighand->siglock);
1323 /* Re-check with the lock held. */
1324 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1325 spin_unlock_irq(&p->sighand->siglock);
1326 return 0;
1327 }
1328 if (!unlikely(wo->wo_flags & WNOWAIT))
1329 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1330 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1331 spin_unlock_irq(&p->sighand->siglock);
1332
1333 pid = task_pid_vnr(p);
1334 get_task_struct(p);
1335 read_unlock(&tasklist_lock);
1336 sched_annotate_sleep();
1337 if (wo->wo_rusage)
1338 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1339 put_task_struct(p);
1340
1341 infop = wo->wo_info;
1342 if (!infop) {
1343 wo->wo_stat = 0xffff;
1344 } else {
1345 infop->cause = CLD_CONTINUED;
1346 infop->pid = pid;
1347 infop->uid = uid;
1348 infop->status = SIGCONT;
1349 }
1350 return pid;
1351}
1352
1353/*
1354 * Consider @p for a wait by @parent.
1355 *
1356 * -ECHILD should be in ->notask_error before the first call.
1357 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1358 * Returns zero if the search for a child should continue;
1359 * then ->notask_error is 0 if @p is an eligible child,
1360 * or still -ECHILD.
1361 */
1362static int wait_consider_task(struct wait_opts *wo, int ptrace,
1363 struct task_struct *p)
1364{
1365 /*
1366 * We can race with wait_task_zombie() from another thread.
1367 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1368 * can't confuse the checks below.
1369 */
1370 int exit_state = READ_ONCE(p->exit_state);
1371 int ret;
1372
1373 if (unlikely(exit_state == EXIT_DEAD))
1374 return 0;
1375
1376 ret = eligible_child(wo, ptrace, p);
1377 if (!ret)
1378 return ret;
1379
1380 if (unlikely(exit_state == EXIT_TRACE)) {
1381 /*
1382 * ptrace == 0 means we are the natural parent. In this case
1383 * we should clear notask_error, debugger will notify us.
1384 */
1385 if (likely(!ptrace))
1386 wo->notask_error = 0;
1387 return 0;
1388 }
1389
1390 if (likely(!ptrace) && unlikely(p->ptrace)) {
1391 /*
1392 * If it is traced by its real parent's group, just pretend
1393 * the caller is ptrace_do_wait() and reap this child if it
1394 * is zombie.
1395 *
1396 * This also hides group stop state from real parent; otherwise
1397 * a single stop can be reported twice as group and ptrace stop.
1398 * If a ptracer wants to distinguish these two events for its
1399 * own children it should create a separate process which takes
1400 * the role of real parent.
1401 */
1402 if (!ptrace_reparented(p))
1403 ptrace = 1;
1404 }
1405
1406 /* slay zombie? */
1407 if (exit_state == EXIT_ZOMBIE) {
1408 /* we don't reap group leaders with subthreads */
1409 if (!delay_group_leader(p)) {
1410 /*
1411 * A zombie ptracee is only visible to its ptracer.
1412 * Notification and reaping will be cascaded to the
1413 * real parent when the ptracer detaches.
1414 */
1415 if (unlikely(ptrace) || likely(!p->ptrace))
1416 return wait_task_zombie(wo, p);
1417 }
1418
1419 /*
1420 * Allow access to stopped/continued state via zombie by
1421 * falling through. Clearing of notask_error is complex.
1422 *
1423 * When !@ptrace:
1424 *
1425 * If WEXITED is set, notask_error should naturally be
1426 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1427 * so, if there are live subthreads, there are events to
1428 * wait for. If all subthreads are dead, it's still safe
1429 * to clear - this function will be called again in finite
1430 * amount time once all the subthreads are released and
1431 * will then return without clearing.
1432 *
1433 * When @ptrace:
1434 *
1435 * Stopped state is per-task and thus can't change once the
1436 * target task dies. Only continued and exited can happen.
1437 * Clear notask_error if WCONTINUED | WEXITED.
1438 */
1439 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1440 wo->notask_error = 0;
1441 } else {
1442 /*
1443 * @p is alive and it's gonna stop, continue or exit, so
1444 * there always is something to wait for.
1445 */
1446 wo->notask_error = 0;
1447 }
1448
1449 /*
1450 * Wait for stopped. Depending on @ptrace, different stopped state
1451 * is used and the two don't interact with each other.
1452 */
1453 ret = wait_task_stopped(wo, ptrace, p);
1454 if (ret)
1455 return ret;
1456
1457 /*
1458 * Wait for continued. There's only one continued state and the
1459 * ptracer can consume it which can confuse the real parent. Don't
1460 * use WCONTINUED from ptracer. You don't need or want it.
1461 */
1462 return wait_task_continued(wo, p);
1463}
1464
1465/*
1466 * Do the work of do_wait() for one thread in the group, @tsk.
1467 *
1468 * -ECHILD should be in ->notask_error before the first call.
1469 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1470 * Returns zero if the search for a child should continue; then
1471 * ->notask_error is 0 if there were any eligible children,
1472 * or still -ECHILD.
1473 */
1474static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1475{
1476 struct task_struct *p;
1477
1478 list_for_each_entry(p, &tsk->children, sibling) {
1479 int ret = wait_consider_task(wo, 0, p);
1480
1481 if (ret)
1482 return ret;
1483 }
1484
1485 return 0;
1486}
1487
1488static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1489{
1490 struct task_struct *p;
1491
1492 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1493 int ret = wait_consider_task(wo, 1, p);
1494
1495 if (ret)
1496 return ret;
1497 }
1498
1499 return 0;
1500}
1501
1502bool pid_child_should_wake(struct wait_opts *wo, struct task_struct *p)
1503{
1504 if (!eligible_pid(wo, p))
1505 return false;
1506
1507 if ((wo->wo_flags & __WNOTHREAD) && wo->child_wait.private != p->parent)
1508 return false;
1509
1510 return true;
1511}
1512
1513static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1514 int sync, void *key)
1515{
1516 struct wait_opts *wo = container_of(wait, struct wait_opts,
1517 child_wait);
1518 struct task_struct *p = key;
1519
1520 if (pid_child_should_wake(wo, p))
1521 return default_wake_function(wait, mode, sync, key);
1522
1523 return 0;
1524}
1525
1526void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1527{
1528 __wake_up_sync_key(&parent->signal->wait_chldexit,
1529 TASK_INTERRUPTIBLE, p);
1530}
1531
1532static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1533 struct task_struct *target)
1534{
1535 struct task_struct *parent =
1536 !ptrace ? target->real_parent : target->parent;
1537
1538 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1539 same_thread_group(current, parent));
1540}
1541
1542/*
1543 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1544 * and tracee lists to find the target task.
1545 */
1546static int do_wait_pid(struct wait_opts *wo)
1547{
1548 bool ptrace;
1549 struct task_struct *target;
1550 int retval;
1551
1552 ptrace = false;
1553 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1554 if (target && is_effectively_child(wo, ptrace, target)) {
1555 retval = wait_consider_task(wo, ptrace, target);
1556 if (retval)
1557 return retval;
1558 }
1559
1560 ptrace = true;
1561 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1562 if (target && target->ptrace &&
1563 is_effectively_child(wo, ptrace, target)) {
1564 retval = wait_consider_task(wo, ptrace, target);
1565 if (retval)
1566 return retval;
1567 }
1568
1569 return 0;
1570}
1571
1572long __do_wait(struct wait_opts *wo)
1573{
1574 long retval;
1575
1576 /*
1577 * If there is nothing that can match our criteria, just get out.
1578 * We will clear ->notask_error to zero if we see any child that
1579 * might later match our criteria, even if we are not able to reap
1580 * it yet.
1581 */
1582 wo->notask_error = -ECHILD;
1583 if ((wo->wo_type < PIDTYPE_MAX) &&
1584 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1585 goto notask;
1586
1587 read_lock(&tasklist_lock);
1588
1589 if (wo->wo_type == PIDTYPE_PID) {
1590 retval = do_wait_pid(wo);
1591 if (retval)
1592 return retval;
1593 } else {
1594 struct task_struct *tsk = current;
1595
1596 do {
1597 retval = do_wait_thread(wo, tsk);
1598 if (retval)
1599 return retval;
1600
1601 retval = ptrace_do_wait(wo, tsk);
1602 if (retval)
1603 return retval;
1604
1605 if (wo->wo_flags & __WNOTHREAD)
1606 break;
1607 } while_each_thread(current, tsk);
1608 }
1609 read_unlock(&tasklist_lock);
1610
1611notask:
1612 retval = wo->notask_error;
1613 if (!retval && !(wo->wo_flags & WNOHANG))
1614 return -ERESTARTSYS;
1615
1616 return retval;
1617}
1618
1619static long do_wait(struct wait_opts *wo)
1620{
1621 int retval;
1622
1623 trace_sched_process_wait(wo->wo_pid);
1624
1625 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1626 wo->child_wait.private = current;
1627 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1628
1629 do {
1630 set_current_state(TASK_INTERRUPTIBLE);
1631 retval = __do_wait(wo);
1632 if (retval != -ERESTARTSYS)
1633 break;
1634 if (signal_pending(current))
1635 break;
1636 schedule();
1637 } while (1);
1638
1639 __set_current_state(TASK_RUNNING);
1640 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1641 return retval;
1642}
1643
1644int kernel_waitid_prepare(struct wait_opts *wo, int which, pid_t upid,
1645 struct waitid_info *infop, int options,
1646 struct rusage *ru)
1647{
1648 unsigned int f_flags = 0;
1649 struct pid *pid = NULL;
1650 enum pid_type type;
1651
1652 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1653 __WNOTHREAD|__WCLONE|__WALL))
1654 return -EINVAL;
1655 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1656 return -EINVAL;
1657
1658 switch (which) {
1659 case P_ALL:
1660 type = PIDTYPE_MAX;
1661 break;
1662 case P_PID:
1663 type = PIDTYPE_PID;
1664 if (upid <= 0)
1665 return -EINVAL;
1666
1667 pid = find_get_pid(upid);
1668 break;
1669 case P_PGID:
1670 type = PIDTYPE_PGID;
1671 if (upid < 0)
1672 return -EINVAL;
1673
1674 if (upid)
1675 pid = find_get_pid(upid);
1676 else
1677 pid = get_task_pid(current, PIDTYPE_PGID);
1678 break;
1679 case P_PIDFD:
1680 type = PIDTYPE_PID;
1681 if (upid < 0)
1682 return -EINVAL;
1683
1684 pid = pidfd_get_pid(upid, &f_flags);
1685 if (IS_ERR(pid))
1686 return PTR_ERR(pid);
1687
1688 break;
1689 default:
1690 return -EINVAL;
1691 }
1692
1693 wo->wo_type = type;
1694 wo->wo_pid = pid;
1695 wo->wo_flags = options;
1696 wo->wo_info = infop;
1697 wo->wo_rusage = ru;
1698 if (f_flags & O_NONBLOCK)
1699 wo->wo_flags |= WNOHANG;
1700
1701 return 0;
1702}
1703
1704static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1705 int options, struct rusage *ru)
1706{
1707 struct wait_opts wo;
1708 long ret;
1709
1710 ret = kernel_waitid_prepare(&wo, which, upid, infop, options, ru);
1711 if (ret)
1712 return ret;
1713
1714 ret = do_wait(&wo);
1715 if (!ret && !(options & WNOHANG) && (wo.wo_flags & WNOHANG))
1716 ret = -EAGAIN;
1717
1718 put_pid(wo.wo_pid);
1719 return ret;
1720}
1721
1722SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1723 infop, int, options, struct rusage __user *, ru)
1724{
1725 struct rusage r;
1726 struct waitid_info info = {.status = 0};
1727 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1728 int signo = 0;
1729
1730 if (err > 0) {
1731 signo = SIGCHLD;
1732 err = 0;
1733 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1734 return -EFAULT;
1735 }
1736 if (!infop)
1737 return err;
1738
1739 if (!user_write_access_begin(infop, sizeof(*infop)))
1740 return -EFAULT;
1741
1742 unsafe_put_user(signo, &infop->si_signo, Efault);
1743 unsafe_put_user(0, &infop->si_errno, Efault);
1744 unsafe_put_user(info.cause, &infop->si_code, Efault);
1745 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1746 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1747 unsafe_put_user(info.status, &infop->si_status, Efault);
1748 user_write_access_end();
1749 return err;
1750Efault:
1751 user_write_access_end();
1752 return -EFAULT;
1753}
1754
1755long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1756 struct rusage *ru)
1757{
1758 struct wait_opts wo;
1759 struct pid *pid = NULL;
1760 enum pid_type type;
1761 long ret;
1762
1763 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1764 __WNOTHREAD|__WCLONE|__WALL))
1765 return -EINVAL;
1766
1767 /* -INT_MIN is not defined */
1768 if (upid == INT_MIN)
1769 return -ESRCH;
1770
1771 if (upid == -1)
1772 type = PIDTYPE_MAX;
1773 else if (upid < 0) {
1774 type = PIDTYPE_PGID;
1775 pid = find_get_pid(-upid);
1776 } else if (upid == 0) {
1777 type = PIDTYPE_PGID;
1778 pid = get_task_pid(current, PIDTYPE_PGID);
1779 } else /* upid > 0 */ {
1780 type = PIDTYPE_PID;
1781 pid = find_get_pid(upid);
1782 }
1783
1784 wo.wo_type = type;
1785 wo.wo_pid = pid;
1786 wo.wo_flags = options | WEXITED;
1787 wo.wo_info = NULL;
1788 wo.wo_stat = 0;
1789 wo.wo_rusage = ru;
1790 ret = do_wait(&wo);
1791 put_pid(pid);
1792 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1793 ret = -EFAULT;
1794
1795 return ret;
1796}
1797
1798int kernel_wait(pid_t pid, int *stat)
1799{
1800 struct wait_opts wo = {
1801 .wo_type = PIDTYPE_PID,
1802 .wo_pid = find_get_pid(pid),
1803 .wo_flags = WEXITED,
1804 };
1805 int ret;
1806
1807 ret = do_wait(&wo);
1808 if (ret > 0 && wo.wo_stat)
1809 *stat = wo.wo_stat;
1810 put_pid(wo.wo_pid);
1811 return ret;
1812}
1813
1814SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1815 int, options, struct rusage __user *, ru)
1816{
1817 struct rusage r;
1818 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1819
1820 if (err > 0) {
1821 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1822 return -EFAULT;
1823 }
1824 return err;
1825}
1826
1827#ifdef __ARCH_WANT_SYS_WAITPID
1828
1829/*
1830 * sys_waitpid() remains for compatibility. waitpid() should be
1831 * implemented by calling sys_wait4() from libc.a.
1832 */
1833SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1834{
1835 return kernel_wait4(pid, stat_addr, options, NULL);
1836}
1837
1838#endif
1839
1840#ifdef CONFIG_COMPAT
1841COMPAT_SYSCALL_DEFINE4(wait4,
1842 compat_pid_t, pid,
1843 compat_uint_t __user *, stat_addr,
1844 int, options,
1845 struct compat_rusage __user *, ru)
1846{
1847 struct rusage r;
1848 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1849 if (err > 0) {
1850 if (ru && put_compat_rusage(&r, ru))
1851 return -EFAULT;
1852 }
1853 return err;
1854}
1855
1856COMPAT_SYSCALL_DEFINE5(waitid,
1857 int, which, compat_pid_t, pid,
1858 struct compat_siginfo __user *, infop, int, options,
1859 struct compat_rusage __user *, uru)
1860{
1861 struct rusage ru;
1862 struct waitid_info info = {.status = 0};
1863 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1864 int signo = 0;
1865 if (err > 0) {
1866 signo = SIGCHLD;
1867 err = 0;
1868 if (uru) {
1869 /* kernel_waitid() overwrites everything in ru */
1870 if (COMPAT_USE_64BIT_TIME)
1871 err = copy_to_user(uru, &ru, sizeof(ru));
1872 else
1873 err = put_compat_rusage(&ru, uru);
1874 if (err)
1875 return -EFAULT;
1876 }
1877 }
1878
1879 if (!infop)
1880 return err;
1881
1882 if (!user_write_access_begin(infop, sizeof(*infop)))
1883 return -EFAULT;
1884
1885 unsafe_put_user(signo, &infop->si_signo, Efault);
1886 unsafe_put_user(0, &infop->si_errno, Efault);
1887 unsafe_put_user(info.cause, &infop->si_code, Efault);
1888 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1889 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1890 unsafe_put_user(info.status, &infop->si_status, Efault);
1891 user_write_access_end();
1892 return err;
1893Efault:
1894 user_write_access_end();
1895 return -EFAULT;
1896}
1897#endif
1898
1899/*
1900 * This needs to be __function_aligned as GCC implicitly makes any
1901 * implementation of abort() cold and drops alignment specified by
1902 * -falign-functions=N.
1903 *
1904 * See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88345#c11
1905 */
1906__weak __function_aligned void abort(void)
1907{
1908 BUG();
1909
1910 /* if that doesn't kill us, halt */
1911 panic("Oops failed to kill thread");
1912}
1913EXPORT_SYMBOL(abort);