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/sched/autogroup.h>
10#include <linux/sched/mm.h>
11#include <linux/sched/stat.h>
12#include <linux/sched/task.h>
13#include <linux/sched/task_stack.h>
14#include <linux/sched/cputime.h>
15#include <linux/interrupt.h>
16#include <linux/module.h>
17#include <linux/capability.h>
18#include <linux/completion.h>
19#include <linux/personality.h>
20#include <linux/tty.h>
21#include <linux/iocontext.h>
22#include <linux/key.h>
23#include <linux/cpu.h>
24#include <linux/acct.h>
25#include <linux/tsacct_kern.h>
26#include <linux/file.h>
27#include <linux/fdtable.h>
28#include <linux/freezer.h>
29#include <linux/binfmts.h>
30#include <linux/nsproxy.h>
31#include <linux/pid_namespace.h>
32#include <linux/ptrace.h>
33#include <linux/profile.h>
34#include <linux/mount.h>
35#include <linux/proc_fs.h>
36#include <linux/kthread.h>
37#include <linux/mempolicy.h>
38#include <linux/taskstats_kern.h>
39#include <linux/delayacct.h>
40#include <linux/cgroup.h>
41#include <linux/syscalls.h>
42#include <linux/signal.h>
43#include <linux/posix-timers.h>
44#include <linux/cn_proc.h>
45#include <linux/mutex.h>
46#include <linux/futex.h>
47#include <linux/pipe_fs_i.h>
48#include <linux/audit.h> /* for audit_free() */
49#include <linux/resource.h>
50#include <linux/blkdev.h>
51#include <linux/task_io_accounting_ops.h>
52#include <linux/tracehook.h>
53#include <linux/fs_struct.h>
54#include <linux/init_task.h>
55#include <linux/perf_event.h>
56#include <trace/events/sched.h>
57#include <linux/hw_breakpoint.h>
58#include <linux/oom.h>
59#include <linux/writeback.h>
60#include <linux/shm.h>
61#include <linux/kcov.h>
62#include <linux/random.h>
63#include <linux/rcuwait.h>
64#include <linux/compat.h>
65
66#include <linux/uaccess.h>
67#include <asm/unistd.h>
68#include <asm/pgtable.h>
69#include <asm/mmu_context.h>
70
71static void __unhash_process(struct task_struct *p, bool group_dead)
72{
73 nr_threads--;
74 detach_pid(p, PIDTYPE_PID);
75 if (group_dead) {
76 detach_pid(p, PIDTYPE_PGID);
77 detach_pid(p, PIDTYPE_SID);
78
79 list_del_rcu(&p->tasks);
80 list_del_init(&p->sibling);
81 __this_cpu_dec(process_counts);
82 }
83 list_del_rcu(&p->thread_group);
84 list_del_rcu(&p->thread_node);
85}
86
87/*
88 * This function expects the tasklist_lock write-locked.
89 */
90static void __exit_signal(struct task_struct *tsk)
91{
92 struct signal_struct *sig = tsk->signal;
93 bool group_dead = thread_group_leader(tsk);
94 struct sighand_struct *sighand;
95 struct tty_struct *uninitialized_var(tty);
96 u64 utime, stime;
97
98 sighand = rcu_dereference_check(tsk->sighand,
99 lockdep_tasklist_lock_is_held());
100 spin_lock(&sighand->siglock);
101
102#ifdef CONFIG_POSIX_TIMERS
103 posix_cpu_timers_exit(tsk);
104 if (group_dead) {
105 posix_cpu_timers_exit_group(tsk);
106 } else {
107 /*
108 * This can only happen if the caller is de_thread().
109 * FIXME: this is the temporary hack, we should teach
110 * posix-cpu-timers to handle this case correctly.
111 */
112 if (unlikely(has_group_leader_pid(tsk)))
113 posix_cpu_timers_exit_group(tsk);
114 }
115#endif
116
117 if (group_dead) {
118 tty = sig->tty;
119 sig->tty = NULL;
120 } else {
121 /*
122 * If there is any task waiting for the group exit
123 * then notify it:
124 */
125 if (sig->notify_count > 0 && !--sig->notify_count)
126 wake_up_process(sig->group_exit_task);
127
128 if (tsk == sig->curr_target)
129 sig->curr_target = next_thread(tsk);
130 }
131
132 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
133 sizeof(unsigned long long));
134
135 /*
136 * Accumulate here the counters for all threads as they die. We could
137 * skip the group leader because it is the last user of signal_struct,
138 * but we want to avoid the race with thread_group_cputime() which can
139 * see the empty ->thread_head list.
140 */
141 task_cputime(tsk, &utime, &stime);
142 write_seqlock(&sig->stats_lock);
143 sig->utime += utime;
144 sig->stime += stime;
145 sig->gtime += task_gtime(tsk);
146 sig->min_flt += tsk->min_flt;
147 sig->maj_flt += tsk->maj_flt;
148 sig->nvcsw += tsk->nvcsw;
149 sig->nivcsw += tsk->nivcsw;
150 sig->inblock += task_io_get_inblock(tsk);
151 sig->oublock += task_io_get_oublock(tsk);
152 task_io_accounting_add(&sig->ioac, &tsk->ioac);
153 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
154 sig->nr_threads--;
155 __unhash_process(tsk, group_dead);
156 write_sequnlock(&sig->stats_lock);
157
158 /*
159 * Do this under ->siglock, we can race with another thread
160 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
161 */
162 flush_sigqueue(&tsk->pending);
163 tsk->sighand = NULL;
164 spin_unlock(&sighand->siglock);
165
166 __cleanup_sighand(sighand);
167 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
168 if (group_dead) {
169 flush_sigqueue(&sig->shared_pending);
170 tty_kref_put(tty);
171 }
172}
173
174static void delayed_put_task_struct(struct rcu_head *rhp)
175{
176 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
177
178 perf_event_delayed_put(tsk);
179 trace_sched_process_free(tsk);
180 put_task_struct(tsk);
181}
182
183
184void release_task(struct task_struct *p)
185{
186 struct task_struct *leader;
187 int zap_leader;
188repeat:
189 /* don't need to get the RCU readlock here - the process is dead and
190 * can't be modifying its own credentials. But shut RCU-lockdep up */
191 rcu_read_lock();
192 atomic_dec(&__task_cred(p)->user->processes);
193 rcu_read_unlock();
194
195 proc_flush_task(p);
196
197 write_lock_irq(&tasklist_lock);
198 ptrace_release_task(p);
199 __exit_signal(p);
200
201 /*
202 * If we are the last non-leader member of the thread
203 * group, and the leader is zombie, then notify the
204 * group leader's parent process. (if it wants notification.)
205 */
206 zap_leader = 0;
207 leader = p->group_leader;
208 if (leader != p && thread_group_empty(leader)
209 && leader->exit_state == EXIT_ZOMBIE) {
210 /*
211 * If we were the last child thread and the leader has
212 * exited already, and the leader's parent ignores SIGCHLD,
213 * then we are the one who should release the leader.
214 */
215 zap_leader = do_notify_parent(leader, leader->exit_signal);
216 if (zap_leader)
217 leader->exit_state = EXIT_DEAD;
218 }
219
220 write_unlock_irq(&tasklist_lock);
221 release_thread(p);
222 call_rcu(&p->rcu, delayed_put_task_struct);
223
224 p = leader;
225 if (unlikely(zap_leader))
226 goto repeat;
227}
228
229/*
230 * Note that if this function returns a valid task_struct pointer (!NULL)
231 * task->usage must remain >0 for the duration of the RCU critical section.
232 */
233struct task_struct *task_rcu_dereference(struct task_struct **ptask)
234{
235 struct sighand_struct *sighand;
236 struct task_struct *task;
237
238 /*
239 * We need to verify that release_task() was not called and thus
240 * delayed_put_task_struct() can't run and drop the last reference
241 * before rcu_read_unlock(). We check task->sighand != NULL,
242 * but we can read the already freed and reused memory.
243 */
244retry:
245 task = rcu_dereference(*ptask);
246 if (!task)
247 return NULL;
248
249 probe_kernel_address(&task->sighand, sighand);
250
251 /*
252 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
253 * was already freed we can not miss the preceding update of this
254 * pointer.
255 */
256 smp_rmb();
257 if (unlikely(task != READ_ONCE(*ptask)))
258 goto retry;
259
260 /*
261 * We've re-checked that "task == *ptask", now we have two different
262 * cases:
263 *
264 * 1. This is actually the same task/task_struct. In this case
265 * sighand != NULL tells us it is still alive.
266 *
267 * 2. This is another task which got the same memory for task_struct.
268 * We can't know this of course, and we can not trust
269 * sighand != NULL.
270 *
271 * In this case we actually return a random value, but this is
272 * correct.
273 *
274 * If we return NULL - we can pretend that we actually noticed that
275 * *ptask was updated when the previous task has exited. Or pretend
276 * that probe_slab_address(&sighand) reads NULL.
277 *
278 * If we return the new task (because sighand is not NULL for any
279 * reason) - this is fine too. This (new) task can't go away before
280 * another gp pass.
281 *
282 * And note: We could even eliminate the false positive if re-read
283 * task->sighand once again to avoid the falsely NULL. But this case
284 * is very unlikely so we don't care.
285 */
286 if (!sighand)
287 return NULL;
288
289 return task;
290}
291
292void rcuwait_wake_up(struct rcuwait *w)
293{
294 struct task_struct *task;
295
296 rcu_read_lock();
297
298 /*
299 * Order condition vs @task, such that everything prior to the load
300 * of @task is visible. This is the condition as to why the user called
301 * rcuwait_trywake() in the first place. Pairs with set_current_state()
302 * barrier (A) in rcuwait_wait_event().
303 *
304 * WAIT WAKE
305 * [S] tsk = current [S] cond = true
306 * MB (A) MB (B)
307 * [L] cond [L] tsk
308 */
309 smp_rmb(); /* (B) */
310
311 /*
312 * Avoid using task_rcu_dereference() magic as long as we are careful,
313 * see comment in rcuwait_wait_event() regarding ->exit_state.
314 */
315 task = rcu_dereference(w->task);
316 if (task)
317 wake_up_process(task);
318 rcu_read_unlock();
319}
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
402#ifdef CONFIG_MEMCG
403/*
404 * A task is exiting. If it owned this mm, find a new owner for the mm.
405 */
406void mm_update_next_owner(struct mm_struct *mm)
407{
408 struct task_struct *c, *g, *p = current;
409
410retry:
411 /*
412 * If the exiting or execing task is not the owner, it's
413 * someone else's problem.
414 */
415 if (mm->owner != p)
416 return;
417 /*
418 * The current owner is exiting/execing and there are no other
419 * candidates. Do not leave the mm pointing to a possibly
420 * freed task structure.
421 */
422 if (atomic_read(&mm->mm_users) <= 1) {
423 mm->owner = NULL;
424 return;
425 }
426
427 read_lock(&tasklist_lock);
428 /*
429 * Search in the children
430 */
431 list_for_each_entry(c, &p->children, sibling) {
432 if (c->mm == mm)
433 goto assign_new_owner;
434 }
435
436 /*
437 * Search in the siblings
438 */
439 list_for_each_entry(c, &p->real_parent->children, sibling) {
440 if (c->mm == mm)
441 goto assign_new_owner;
442 }
443
444 /*
445 * Search through everything else, we should not get here often.
446 */
447 for_each_process(g) {
448 if (g->flags & PF_KTHREAD)
449 continue;
450 for_each_thread(g, c) {
451 if (c->mm == mm)
452 goto assign_new_owner;
453 if (c->mm)
454 break;
455 }
456 }
457 read_unlock(&tasklist_lock);
458 /*
459 * We found no owner yet mm_users > 1: this implies that we are
460 * most likely racing with swapoff (try_to_unuse()) or /proc or
461 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
462 */
463 mm->owner = NULL;
464 return;
465
466assign_new_owner:
467 BUG_ON(c == p);
468 get_task_struct(c);
469 /*
470 * The task_lock protects c->mm from changing.
471 * We always want mm->owner->mm == mm
472 */
473 task_lock(c);
474 /*
475 * Delay read_unlock() till we have the task_lock()
476 * to ensure that c does not slip away underneath us
477 */
478 read_unlock(&tasklist_lock);
479 if (c->mm != mm) {
480 task_unlock(c);
481 put_task_struct(c);
482 goto retry;
483 }
484 mm->owner = c;
485 task_unlock(c);
486 put_task_struct(c);
487}
488#endif /* CONFIG_MEMCG */
489
490/*
491 * Turn us into a lazy TLB process if we
492 * aren't already..
493 */
494static void exit_mm(void)
495{
496 struct mm_struct *mm = current->mm;
497 struct core_state *core_state;
498
499 mm_release(current, mm);
500 if (!mm)
501 return;
502 sync_mm_rss(mm);
503 /*
504 * Serialize with any possible pending coredump.
505 * We must hold mmap_sem around checking core_state
506 * and clearing tsk->mm. The core-inducing thread
507 * will increment ->nr_threads for each thread in the
508 * group with ->mm != NULL.
509 */
510 down_read(&mm->mmap_sem);
511 core_state = mm->core_state;
512 if (core_state) {
513 struct core_thread self;
514
515 up_read(&mm->mmap_sem);
516
517 self.task = current;
518 self.next = xchg(&core_state->dumper.next, &self);
519 /*
520 * Implies mb(), the result of xchg() must be visible
521 * to core_state->dumper.
522 */
523 if (atomic_dec_and_test(&core_state->nr_threads))
524 complete(&core_state->startup);
525
526 for (;;) {
527 set_current_state(TASK_UNINTERRUPTIBLE);
528 if (!self.task) /* see coredump_finish() */
529 break;
530 freezable_schedule();
531 }
532 __set_current_state(TASK_RUNNING);
533 down_read(&mm->mmap_sem);
534 }
535 mmgrab(mm);
536 BUG_ON(mm != current->active_mm);
537 /* more a memory barrier than a real lock */
538 task_lock(current);
539 current->mm = NULL;
540 up_read(&mm->mmap_sem);
541 enter_lazy_tlb(mm, current);
542 task_unlock(current);
543 mm_update_next_owner(mm);
544 mmput(mm);
545 if (test_thread_flag(TIF_MEMDIE))
546 exit_oom_victim();
547}
548
549static struct task_struct *find_alive_thread(struct task_struct *p)
550{
551 struct task_struct *t;
552
553 for_each_thread(p, t) {
554 if (!(t->flags & PF_EXITING))
555 return t;
556 }
557 return NULL;
558}
559
560static struct task_struct *find_child_reaper(struct task_struct *father)
561 __releases(&tasklist_lock)
562 __acquires(&tasklist_lock)
563{
564 struct pid_namespace *pid_ns = task_active_pid_ns(father);
565 struct task_struct *reaper = pid_ns->child_reaper;
566
567 if (likely(reaper != father))
568 return reaper;
569
570 reaper = find_alive_thread(father);
571 if (reaper) {
572 pid_ns->child_reaper = reaper;
573 return reaper;
574 }
575
576 write_unlock_irq(&tasklist_lock);
577 if (unlikely(pid_ns == &init_pid_ns)) {
578 panic("Attempted to kill init! exitcode=0x%08x\n",
579 father->signal->group_exit_code ?: father->exit_code);
580 }
581 zap_pid_ns_processes(pid_ns);
582 write_lock_irq(&tasklist_lock);
583
584 return father;
585}
586
587/*
588 * When we die, we re-parent all our children, and try to:
589 * 1. give them to another thread in our thread group, if such a member exists
590 * 2. give it to the first ancestor process which prctl'd itself as a
591 * child_subreaper for its children (like a service manager)
592 * 3. give it to the init process (PID 1) in our pid namespace
593 */
594static struct task_struct *find_new_reaper(struct task_struct *father,
595 struct task_struct *child_reaper)
596{
597 struct task_struct *thread, *reaper;
598
599 thread = find_alive_thread(father);
600 if (thread)
601 return thread;
602
603 if (father->signal->has_child_subreaper) {
604 unsigned int ns_level = task_pid(father)->level;
605 /*
606 * Find the first ->is_child_subreaper ancestor in our pid_ns.
607 * We can't check reaper != child_reaper to ensure we do not
608 * cross the namespaces, the exiting parent could be injected
609 * by setns() + fork().
610 * We check pid->level, this is slightly more efficient than
611 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
612 */
613 for (reaper = father->real_parent;
614 task_pid(reaper)->level == ns_level;
615 reaper = reaper->real_parent) {
616 if (reaper == &init_task)
617 break;
618 if (!reaper->signal->is_child_subreaper)
619 continue;
620 thread = find_alive_thread(reaper);
621 if (thread)
622 return thread;
623 }
624 }
625
626 return child_reaper;
627}
628
629/*
630* Any that need to be release_task'd are put on the @dead list.
631 */
632static void reparent_leader(struct task_struct *father, struct task_struct *p,
633 struct list_head *dead)
634{
635 if (unlikely(p->exit_state == EXIT_DEAD))
636 return;
637
638 /* We don't want people slaying init. */
639 p->exit_signal = SIGCHLD;
640
641 /* If it has exited notify the new parent about this child's death. */
642 if (!p->ptrace &&
643 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
644 if (do_notify_parent(p, p->exit_signal)) {
645 p->exit_state = EXIT_DEAD;
646 list_add(&p->ptrace_entry, dead);
647 }
648 }
649
650 kill_orphaned_pgrp(p, father);
651}
652
653/*
654 * This does two things:
655 *
656 * A. Make init inherit all the child processes
657 * B. Check to see if any process groups have become orphaned
658 * as a result of our exiting, and if they have any stopped
659 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
660 */
661static void forget_original_parent(struct task_struct *father,
662 struct list_head *dead)
663{
664 struct task_struct *p, *t, *reaper;
665
666 if (unlikely(!list_empty(&father->ptraced)))
667 exit_ptrace(father, dead);
668
669 /* Can drop and reacquire tasklist_lock */
670 reaper = find_child_reaper(father);
671 if (list_empty(&father->children))
672 return;
673
674 reaper = find_new_reaper(father, reaper);
675 list_for_each_entry(p, &father->children, sibling) {
676 for_each_thread(p, t) {
677 t->real_parent = reaper;
678 BUG_ON((!t->ptrace) != (t->parent == father));
679 if (likely(!t->ptrace))
680 t->parent = t->real_parent;
681 if (t->pdeath_signal)
682 group_send_sig_info(t->pdeath_signal,
683 SEND_SIG_NOINFO, t);
684 }
685 /*
686 * If this is a threaded reparent there is no need to
687 * notify anyone anything has happened.
688 */
689 if (!same_thread_group(reaper, father))
690 reparent_leader(father, p, dead);
691 }
692 list_splice_tail_init(&father->children, &reaper->children);
693}
694
695/*
696 * Send signals to all our closest relatives so that they know
697 * to properly mourn us..
698 */
699static void exit_notify(struct task_struct *tsk, int group_dead)
700{
701 bool autoreap;
702 struct task_struct *p, *n;
703 LIST_HEAD(dead);
704
705 write_lock_irq(&tasklist_lock);
706 forget_original_parent(tsk, &dead);
707
708 if (group_dead)
709 kill_orphaned_pgrp(tsk->group_leader, NULL);
710
711 if (unlikely(tsk->ptrace)) {
712 int sig = thread_group_leader(tsk) &&
713 thread_group_empty(tsk) &&
714 !ptrace_reparented(tsk) ?
715 tsk->exit_signal : SIGCHLD;
716 autoreap = do_notify_parent(tsk, sig);
717 } else if (thread_group_leader(tsk)) {
718 autoreap = thread_group_empty(tsk) &&
719 do_notify_parent(tsk, tsk->exit_signal);
720 } else {
721 autoreap = true;
722 }
723
724 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
725 if (tsk->exit_state == EXIT_DEAD)
726 list_add(&tsk->ptrace_entry, &dead);
727
728 /* mt-exec, de_thread() is waiting for group leader */
729 if (unlikely(tsk->signal->notify_count < 0))
730 wake_up_process(tsk->signal->group_exit_task);
731 write_unlock_irq(&tasklist_lock);
732
733 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
734 list_del_init(&p->ptrace_entry);
735 release_task(p);
736 }
737}
738
739#ifdef CONFIG_DEBUG_STACK_USAGE
740static void check_stack_usage(void)
741{
742 static DEFINE_SPINLOCK(low_water_lock);
743 static int lowest_to_date = THREAD_SIZE;
744 unsigned long free;
745
746 free = stack_not_used(current);
747
748 if (free >= lowest_to_date)
749 return;
750
751 spin_lock(&low_water_lock);
752 if (free < lowest_to_date) {
753 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
754 current->comm, task_pid_nr(current), free);
755 lowest_to_date = free;
756 }
757 spin_unlock(&low_water_lock);
758}
759#else
760static inline void check_stack_usage(void) {}
761#endif
762
763void __noreturn do_exit(long code)
764{
765 struct task_struct *tsk = current;
766 int group_dead;
767
768 profile_task_exit(tsk);
769 kcov_task_exit(tsk);
770
771 WARN_ON(blk_needs_flush_plug(tsk));
772
773 if (unlikely(in_interrupt()))
774 panic("Aiee, killing interrupt handler!");
775 if (unlikely(!tsk->pid))
776 panic("Attempted to kill the idle task!");
777
778 /*
779 * If do_exit is called because this processes oopsed, it's possible
780 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
781 * continuing. Amongst other possible reasons, this is to prevent
782 * mm_release()->clear_child_tid() from writing to a user-controlled
783 * kernel address.
784 */
785 set_fs(USER_DS);
786
787 ptrace_event(PTRACE_EVENT_EXIT, code);
788
789 validate_creds_for_do_exit(tsk);
790
791 /*
792 * We're taking recursive faults here in do_exit. Safest is to just
793 * leave this task alone and wait for reboot.
794 */
795 if (unlikely(tsk->flags & PF_EXITING)) {
796 pr_alert("Fixing recursive fault but reboot is needed!\n");
797 /*
798 * We can do this unlocked here. The futex code uses
799 * this flag just to verify whether the pi state
800 * cleanup has been done or not. In the worst case it
801 * loops once more. We pretend that the cleanup was
802 * done as there is no way to return. Either the
803 * OWNER_DIED bit is set by now or we push the blocked
804 * task into the wait for ever nirwana as well.
805 */
806 tsk->flags |= PF_EXITPIDONE;
807 set_current_state(TASK_UNINTERRUPTIBLE);
808 schedule();
809 }
810
811 exit_signals(tsk); /* sets PF_EXITING */
812 /*
813 * Ensure that all new tsk->pi_lock acquisitions must observe
814 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
815 */
816 smp_mb();
817 /*
818 * Ensure that we must observe the pi_state in exit_mm() ->
819 * mm_release() -> exit_pi_state_list().
820 */
821 raw_spin_lock_irq(&tsk->pi_lock);
822 raw_spin_unlock_irq(&tsk->pi_lock);
823
824 if (unlikely(in_atomic())) {
825 pr_info("note: %s[%d] exited with preempt_count %d\n",
826 current->comm, task_pid_nr(current),
827 preempt_count());
828 preempt_count_set(PREEMPT_ENABLED);
829 }
830
831 /* sync mm's RSS info before statistics gathering */
832 if (tsk->mm)
833 sync_mm_rss(tsk->mm);
834 acct_update_integrals(tsk);
835 group_dead = atomic_dec_and_test(&tsk->signal->live);
836 if (group_dead) {
837#ifdef CONFIG_POSIX_TIMERS
838 hrtimer_cancel(&tsk->signal->real_timer);
839 exit_itimers(tsk->signal);
840#endif
841 if (tsk->mm)
842 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
843 }
844 acct_collect(code, group_dead);
845 if (group_dead)
846 tty_audit_exit();
847 audit_free(tsk);
848
849 tsk->exit_code = code;
850 taskstats_exit(tsk, group_dead);
851
852 exit_mm();
853
854 if (group_dead)
855 acct_process();
856 trace_sched_process_exit(tsk);
857
858 exit_sem(tsk);
859 exit_shm(tsk);
860 exit_files(tsk);
861 exit_fs(tsk);
862 if (group_dead)
863 disassociate_ctty(1);
864 exit_task_namespaces(tsk);
865 exit_task_work(tsk);
866 exit_thread(tsk);
867
868 /*
869 * Flush inherited counters to the parent - before the parent
870 * gets woken up by child-exit notifications.
871 *
872 * because of cgroup mode, must be called before cgroup_exit()
873 */
874 perf_event_exit_task(tsk);
875
876 sched_autogroup_exit_task(tsk);
877 cgroup_exit(tsk);
878
879 /*
880 * FIXME: do that only when needed, using sched_exit tracepoint
881 */
882 flush_ptrace_hw_breakpoint(tsk);
883
884 exit_tasks_rcu_start();
885 exit_notify(tsk, group_dead);
886 proc_exit_connector(tsk);
887 mpol_put_task_policy(tsk);
888#ifdef CONFIG_FUTEX
889 if (unlikely(current->pi_state_cache))
890 kfree(current->pi_state_cache);
891#endif
892 /*
893 * Make sure we are holding no locks:
894 */
895 debug_check_no_locks_held();
896 /*
897 * We can do this unlocked here. The futex code uses this flag
898 * just to verify whether the pi state cleanup has been done
899 * or not. In the worst case it loops once more.
900 */
901 tsk->flags |= PF_EXITPIDONE;
902
903 if (tsk->io_context)
904 exit_io_context(tsk);
905
906 if (tsk->splice_pipe)
907 free_pipe_info(tsk->splice_pipe);
908
909 if (tsk->task_frag.page)
910 put_page(tsk->task_frag.page);
911
912 validate_creds_for_do_exit(tsk);
913
914 check_stack_usage();
915 preempt_disable();
916 if (tsk->nr_dirtied)
917 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
918 exit_rcu();
919 exit_tasks_rcu_finish();
920
921 lockdep_free_task(tsk);
922 do_task_dead();
923}
924EXPORT_SYMBOL_GPL(do_exit);
925
926void complete_and_exit(struct completion *comp, long code)
927{
928 if (comp)
929 complete(comp);
930
931 do_exit(code);
932}
933EXPORT_SYMBOL(complete_and_exit);
934
935SYSCALL_DEFINE1(exit, int, error_code)
936{
937 do_exit((error_code&0xff)<<8);
938}
939
940/*
941 * Take down every thread in the group. This is called by fatal signals
942 * as well as by sys_exit_group (below).
943 */
944void
945do_group_exit(int exit_code)
946{
947 struct signal_struct *sig = current->signal;
948
949 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
950
951 if (signal_group_exit(sig))
952 exit_code = sig->group_exit_code;
953 else if (!thread_group_empty(current)) {
954 struct sighand_struct *const sighand = current->sighand;
955
956 spin_lock_irq(&sighand->siglock);
957 if (signal_group_exit(sig))
958 /* Another thread got here before we took the lock. */
959 exit_code = sig->group_exit_code;
960 else {
961 sig->group_exit_code = exit_code;
962 sig->flags = SIGNAL_GROUP_EXIT;
963 zap_other_threads(current);
964 }
965 spin_unlock_irq(&sighand->siglock);
966 }
967
968 do_exit(exit_code);
969 /* NOTREACHED */
970}
971
972/*
973 * this kills every thread in the thread group. Note that any externally
974 * wait4()-ing process will get the correct exit code - even if this
975 * thread is not the thread group leader.
976 */
977SYSCALL_DEFINE1(exit_group, int, error_code)
978{
979 do_group_exit((error_code & 0xff) << 8);
980 /* NOTREACHED */
981 return 0;
982}
983
984struct waitid_info {
985 pid_t pid;
986 uid_t uid;
987 int status;
988 int cause;
989};
990
991struct wait_opts {
992 enum pid_type wo_type;
993 int wo_flags;
994 struct pid *wo_pid;
995
996 struct waitid_info *wo_info;
997 int wo_stat;
998 struct rusage *wo_rusage;
999
1000 wait_queue_entry_t child_wait;
1001 int notask_error;
1002};
1003
1004static inline
1005struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1006{
1007 if (type != PIDTYPE_PID)
1008 task = task->group_leader;
1009 return task->pids[type].pid;
1010}
1011
1012static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1013{
1014 return wo->wo_type == PIDTYPE_MAX ||
1015 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1016}
1017
1018static int
1019eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1020{
1021 if (!eligible_pid(wo, p))
1022 return 0;
1023
1024 /*
1025 * Wait for all children (clone and not) if __WALL is set or
1026 * if it is traced by us.
1027 */
1028 if (ptrace || (wo->wo_flags & __WALL))
1029 return 1;
1030
1031 /*
1032 * Otherwise, wait for clone children *only* if __WCLONE is set;
1033 * otherwise, wait for non-clone children *only*.
1034 *
1035 * Note: a "clone" child here is one that reports to its parent
1036 * using a signal other than SIGCHLD, or a non-leader thread which
1037 * we can only see if it is traced by us.
1038 */
1039 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1040 return 0;
1041
1042 return 1;
1043}
1044
1045/*
1046 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1047 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1048 * the lock and this task is uninteresting. If we return nonzero, we have
1049 * released the lock and the system call should return.
1050 */
1051static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1052{
1053 int state, status;
1054 pid_t pid = task_pid_vnr(p);
1055 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1056 struct waitid_info *infop;
1057
1058 if (!likely(wo->wo_flags & WEXITED))
1059 return 0;
1060
1061 if (unlikely(wo->wo_flags & WNOWAIT)) {
1062 status = p->exit_code;
1063 get_task_struct(p);
1064 read_unlock(&tasklist_lock);
1065 sched_annotate_sleep();
1066 if (wo->wo_rusage)
1067 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1068 put_task_struct(p);
1069 goto out_info;
1070 }
1071 /*
1072 * Move the task's state to DEAD/TRACE, only one thread can do this.
1073 */
1074 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1075 EXIT_TRACE : EXIT_DEAD;
1076 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1077 return 0;
1078 /*
1079 * We own this thread, nobody else can reap it.
1080 */
1081 read_unlock(&tasklist_lock);
1082 sched_annotate_sleep();
1083
1084 /*
1085 * Check thread_group_leader() to exclude the traced sub-threads.
1086 */
1087 if (state == EXIT_DEAD && thread_group_leader(p)) {
1088 struct signal_struct *sig = p->signal;
1089 struct signal_struct *psig = current->signal;
1090 unsigned long maxrss;
1091 u64 tgutime, tgstime;
1092
1093 /*
1094 * The resource counters for the group leader are in its
1095 * own task_struct. Those for dead threads in the group
1096 * are in its signal_struct, as are those for the child
1097 * processes it has previously reaped. All these
1098 * accumulate in the parent's signal_struct c* fields.
1099 *
1100 * We don't bother to take a lock here to protect these
1101 * p->signal fields because the whole thread group is dead
1102 * and nobody can change them.
1103 *
1104 * psig->stats_lock also protects us from our sub-theads
1105 * which can reap other children at the same time. Until
1106 * we change k_getrusage()-like users to rely on this lock
1107 * we have to take ->siglock as well.
1108 *
1109 * We use thread_group_cputime_adjusted() to get times for
1110 * the thread group, which consolidates times for all threads
1111 * in the group including the group leader.
1112 */
1113 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1114 spin_lock_irq(¤t->sighand->siglock);
1115 write_seqlock(&psig->stats_lock);
1116 psig->cutime += tgutime + sig->cutime;
1117 psig->cstime += tgstime + sig->cstime;
1118 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1119 psig->cmin_flt +=
1120 p->min_flt + sig->min_flt + sig->cmin_flt;
1121 psig->cmaj_flt +=
1122 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1123 psig->cnvcsw +=
1124 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1125 psig->cnivcsw +=
1126 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1127 psig->cinblock +=
1128 task_io_get_inblock(p) +
1129 sig->inblock + sig->cinblock;
1130 psig->coublock +=
1131 task_io_get_oublock(p) +
1132 sig->oublock + sig->coublock;
1133 maxrss = max(sig->maxrss, sig->cmaxrss);
1134 if (psig->cmaxrss < maxrss)
1135 psig->cmaxrss = maxrss;
1136 task_io_accounting_add(&psig->ioac, &p->ioac);
1137 task_io_accounting_add(&psig->ioac, &sig->ioac);
1138 write_sequnlock(&psig->stats_lock);
1139 spin_unlock_irq(¤t->sighand->siglock);
1140 }
1141
1142 if (wo->wo_rusage)
1143 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1144 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1145 ? p->signal->group_exit_code : p->exit_code;
1146 wo->wo_stat = status;
1147
1148 if (state == EXIT_TRACE) {
1149 write_lock_irq(&tasklist_lock);
1150 /* We dropped tasklist, ptracer could die and untrace */
1151 ptrace_unlink(p);
1152
1153 /* If parent wants a zombie, don't release it now */
1154 state = EXIT_ZOMBIE;
1155 if (do_notify_parent(p, p->exit_signal))
1156 state = EXIT_DEAD;
1157 p->exit_state = state;
1158 write_unlock_irq(&tasklist_lock);
1159 }
1160 if (state == EXIT_DEAD)
1161 release_task(p);
1162
1163out_info:
1164 infop = wo->wo_info;
1165 if (infop) {
1166 if ((status & 0x7f) == 0) {
1167 infop->cause = CLD_EXITED;
1168 infop->status = status >> 8;
1169 } else {
1170 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1171 infop->status = status & 0x7f;
1172 }
1173 infop->pid = pid;
1174 infop->uid = uid;
1175 }
1176
1177 return pid;
1178}
1179
1180static int *task_stopped_code(struct task_struct *p, bool ptrace)
1181{
1182 if (ptrace) {
1183 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1184 return &p->exit_code;
1185 } else {
1186 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1187 return &p->signal->group_exit_code;
1188 }
1189 return NULL;
1190}
1191
1192/**
1193 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1194 * @wo: wait options
1195 * @ptrace: is the wait for ptrace
1196 * @p: task to wait for
1197 *
1198 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1199 *
1200 * CONTEXT:
1201 * read_lock(&tasklist_lock), which is released if return value is
1202 * non-zero. Also, grabs and releases @p->sighand->siglock.
1203 *
1204 * RETURNS:
1205 * 0 if wait condition didn't exist and search for other wait conditions
1206 * should continue. Non-zero return, -errno on failure and @p's pid on
1207 * success, implies that tasklist_lock is released and wait condition
1208 * search should terminate.
1209 */
1210static int wait_task_stopped(struct wait_opts *wo,
1211 int ptrace, struct task_struct *p)
1212{
1213 struct waitid_info *infop;
1214 int exit_code, *p_code, why;
1215 uid_t uid = 0; /* unneeded, required by compiler */
1216 pid_t pid;
1217
1218 /*
1219 * Traditionally we see ptrace'd stopped tasks regardless of options.
1220 */
1221 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1222 return 0;
1223
1224 if (!task_stopped_code(p, ptrace))
1225 return 0;
1226
1227 exit_code = 0;
1228 spin_lock_irq(&p->sighand->siglock);
1229
1230 p_code = task_stopped_code(p, ptrace);
1231 if (unlikely(!p_code))
1232 goto unlock_sig;
1233
1234 exit_code = *p_code;
1235 if (!exit_code)
1236 goto unlock_sig;
1237
1238 if (!unlikely(wo->wo_flags & WNOWAIT))
1239 *p_code = 0;
1240
1241 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1242unlock_sig:
1243 spin_unlock_irq(&p->sighand->siglock);
1244 if (!exit_code)
1245 return 0;
1246
1247 /*
1248 * Now we are pretty sure this task is interesting.
1249 * Make sure it doesn't get reaped out from under us while we
1250 * give up the lock and then examine it below. We don't want to
1251 * keep holding onto the tasklist_lock while we call getrusage and
1252 * possibly take page faults for user memory.
1253 */
1254 get_task_struct(p);
1255 pid = task_pid_vnr(p);
1256 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1257 read_unlock(&tasklist_lock);
1258 sched_annotate_sleep();
1259 if (wo->wo_rusage)
1260 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1261 put_task_struct(p);
1262
1263 if (likely(!(wo->wo_flags & WNOWAIT)))
1264 wo->wo_stat = (exit_code << 8) | 0x7f;
1265
1266 infop = wo->wo_info;
1267 if (infop) {
1268 infop->cause = why;
1269 infop->status = exit_code;
1270 infop->pid = pid;
1271 infop->uid = uid;
1272 }
1273 return pid;
1274}
1275
1276/*
1277 * Handle do_wait work for one task in a live, non-stopped state.
1278 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1279 * the lock and this task is uninteresting. If we return nonzero, we have
1280 * released the lock and the system call should return.
1281 */
1282static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1283{
1284 struct waitid_info *infop;
1285 pid_t pid;
1286 uid_t uid;
1287
1288 if (!unlikely(wo->wo_flags & WCONTINUED))
1289 return 0;
1290
1291 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1292 return 0;
1293
1294 spin_lock_irq(&p->sighand->siglock);
1295 /* Re-check with the lock held. */
1296 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1297 spin_unlock_irq(&p->sighand->siglock);
1298 return 0;
1299 }
1300 if (!unlikely(wo->wo_flags & WNOWAIT))
1301 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1302 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1303 spin_unlock_irq(&p->sighand->siglock);
1304
1305 pid = task_pid_vnr(p);
1306 get_task_struct(p);
1307 read_unlock(&tasklist_lock);
1308 sched_annotate_sleep();
1309 if (wo->wo_rusage)
1310 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1311 put_task_struct(p);
1312
1313 infop = wo->wo_info;
1314 if (!infop) {
1315 wo->wo_stat = 0xffff;
1316 } else {
1317 infop->cause = CLD_CONTINUED;
1318 infop->pid = pid;
1319 infop->uid = uid;
1320 infop->status = SIGCONT;
1321 }
1322 return pid;
1323}
1324
1325/*
1326 * Consider @p for a wait by @parent.
1327 *
1328 * -ECHILD should be in ->notask_error before the first call.
1329 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1330 * Returns zero if the search for a child should continue;
1331 * then ->notask_error is 0 if @p is an eligible child,
1332 * or still -ECHILD.
1333 */
1334static int wait_consider_task(struct wait_opts *wo, int ptrace,
1335 struct task_struct *p)
1336{
1337 /*
1338 * We can race with wait_task_zombie() from another thread.
1339 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1340 * can't confuse the checks below.
1341 */
1342 int exit_state = READ_ONCE(p->exit_state);
1343 int ret;
1344
1345 if (unlikely(exit_state == EXIT_DEAD))
1346 return 0;
1347
1348 ret = eligible_child(wo, ptrace, p);
1349 if (!ret)
1350 return ret;
1351
1352 if (unlikely(exit_state == EXIT_TRACE)) {
1353 /*
1354 * ptrace == 0 means we are the natural parent. In this case
1355 * we should clear notask_error, debugger will notify us.
1356 */
1357 if (likely(!ptrace))
1358 wo->notask_error = 0;
1359 return 0;
1360 }
1361
1362 if (likely(!ptrace) && unlikely(p->ptrace)) {
1363 /*
1364 * If it is traced by its real parent's group, just pretend
1365 * the caller is ptrace_do_wait() and reap this child if it
1366 * is zombie.
1367 *
1368 * This also hides group stop state from real parent; otherwise
1369 * a single stop can be reported twice as group and ptrace stop.
1370 * If a ptracer wants to distinguish these two events for its
1371 * own children it should create a separate process which takes
1372 * the role of real parent.
1373 */
1374 if (!ptrace_reparented(p))
1375 ptrace = 1;
1376 }
1377
1378 /* slay zombie? */
1379 if (exit_state == EXIT_ZOMBIE) {
1380 /* we don't reap group leaders with subthreads */
1381 if (!delay_group_leader(p)) {
1382 /*
1383 * A zombie ptracee is only visible to its ptracer.
1384 * Notification and reaping will be cascaded to the
1385 * real parent when the ptracer detaches.
1386 */
1387 if (unlikely(ptrace) || likely(!p->ptrace))
1388 return wait_task_zombie(wo, p);
1389 }
1390
1391 /*
1392 * Allow access to stopped/continued state via zombie by
1393 * falling through. Clearing of notask_error is complex.
1394 *
1395 * When !@ptrace:
1396 *
1397 * If WEXITED is set, notask_error should naturally be
1398 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1399 * so, if there are live subthreads, there are events to
1400 * wait for. If all subthreads are dead, it's still safe
1401 * to clear - this function will be called again in finite
1402 * amount time once all the subthreads are released and
1403 * will then return without clearing.
1404 *
1405 * When @ptrace:
1406 *
1407 * Stopped state is per-task and thus can't change once the
1408 * target task dies. Only continued and exited can happen.
1409 * Clear notask_error if WCONTINUED | WEXITED.
1410 */
1411 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1412 wo->notask_error = 0;
1413 } else {
1414 /*
1415 * @p is alive and it's gonna stop, continue or exit, so
1416 * there always is something to wait for.
1417 */
1418 wo->notask_error = 0;
1419 }
1420
1421 /*
1422 * Wait for stopped. Depending on @ptrace, different stopped state
1423 * is used and the two don't interact with each other.
1424 */
1425 ret = wait_task_stopped(wo, ptrace, p);
1426 if (ret)
1427 return ret;
1428
1429 /*
1430 * Wait for continued. There's only one continued state and the
1431 * ptracer can consume it which can confuse the real parent. Don't
1432 * use WCONTINUED from ptracer. You don't need or want it.
1433 */
1434 return wait_task_continued(wo, p);
1435}
1436
1437/*
1438 * Do the work of do_wait() for one thread in the group, @tsk.
1439 *
1440 * -ECHILD should be in ->notask_error before the first call.
1441 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1442 * Returns zero if the search for a child should continue; then
1443 * ->notask_error is 0 if there were any eligible children,
1444 * or still -ECHILD.
1445 */
1446static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1447{
1448 struct task_struct *p;
1449
1450 list_for_each_entry(p, &tsk->children, sibling) {
1451 int ret = wait_consider_task(wo, 0, p);
1452
1453 if (ret)
1454 return ret;
1455 }
1456
1457 return 0;
1458}
1459
1460static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1461{
1462 struct task_struct *p;
1463
1464 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1465 int ret = wait_consider_task(wo, 1, p);
1466
1467 if (ret)
1468 return ret;
1469 }
1470
1471 return 0;
1472}
1473
1474static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1475 int sync, void *key)
1476{
1477 struct wait_opts *wo = container_of(wait, struct wait_opts,
1478 child_wait);
1479 struct task_struct *p = key;
1480
1481 if (!eligible_pid(wo, p))
1482 return 0;
1483
1484 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1485 return 0;
1486
1487 return default_wake_function(wait, mode, sync, key);
1488}
1489
1490void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1491{
1492 __wake_up_sync_key(&parent->signal->wait_chldexit,
1493 TASK_INTERRUPTIBLE, 1, p);
1494}
1495
1496static long do_wait(struct wait_opts *wo)
1497{
1498 struct task_struct *tsk;
1499 int retval;
1500
1501 trace_sched_process_wait(wo->wo_pid);
1502
1503 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1504 wo->child_wait.private = current;
1505 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1506repeat:
1507 /*
1508 * If there is nothing that can match our criteria, just get out.
1509 * We will clear ->notask_error to zero if we see any child that
1510 * might later match our criteria, even if we are not able to reap
1511 * it yet.
1512 */
1513 wo->notask_error = -ECHILD;
1514 if ((wo->wo_type < PIDTYPE_MAX) &&
1515 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1516 goto notask;
1517
1518 set_current_state(TASK_INTERRUPTIBLE);
1519 read_lock(&tasklist_lock);
1520 tsk = current;
1521 do {
1522 retval = do_wait_thread(wo, tsk);
1523 if (retval)
1524 goto end;
1525
1526 retval = ptrace_do_wait(wo, tsk);
1527 if (retval)
1528 goto end;
1529
1530 if (wo->wo_flags & __WNOTHREAD)
1531 break;
1532 } while_each_thread(current, tsk);
1533 read_unlock(&tasklist_lock);
1534
1535notask:
1536 retval = wo->notask_error;
1537 if (!retval && !(wo->wo_flags & WNOHANG)) {
1538 retval = -ERESTARTSYS;
1539 if (!signal_pending(current)) {
1540 schedule();
1541 goto repeat;
1542 }
1543 }
1544end:
1545 __set_current_state(TASK_RUNNING);
1546 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1547 return retval;
1548}
1549
1550static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1551 int options, struct rusage *ru)
1552{
1553 struct wait_opts wo;
1554 struct pid *pid = NULL;
1555 enum pid_type type;
1556 long ret;
1557
1558 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1559 __WNOTHREAD|__WCLONE|__WALL))
1560 return -EINVAL;
1561 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1562 return -EINVAL;
1563
1564 switch (which) {
1565 case P_ALL:
1566 type = PIDTYPE_MAX;
1567 break;
1568 case P_PID:
1569 type = PIDTYPE_PID;
1570 if (upid <= 0)
1571 return -EINVAL;
1572 break;
1573 case P_PGID:
1574 type = PIDTYPE_PGID;
1575 if (upid <= 0)
1576 return -EINVAL;
1577 break;
1578 default:
1579 return -EINVAL;
1580 }
1581
1582 if (type < PIDTYPE_MAX)
1583 pid = find_get_pid(upid);
1584
1585 wo.wo_type = type;
1586 wo.wo_pid = pid;
1587 wo.wo_flags = options;
1588 wo.wo_info = infop;
1589 wo.wo_rusage = ru;
1590 ret = do_wait(&wo);
1591
1592 put_pid(pid);
1593 return ret;
1594}
1595
1596SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1597 infop, int, options, struct rusage __user *, ru)
1598{
1599 struct rusage r;
1600 struct waitid_info info = {.status = 0};
1601 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1602 int signo = 0;
1603
1604 if (err > 0) {
1605 signo = SIGCHLD;
1606 err = 0;
1607 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1608 return -EFAULT;
1609 }
1610 if (!infop)
1611 return err;
1612
1613 if (!access_ok(VERIFY_WRITE, infop, sizeof(*infop)))
1614 return -EFAULT;
1615
1616 user_access_begin();
1617 unsafe_put_user(signo, &infop->si_signo, Efault);
1618 unsafe_put_user(0, &infop->si_errno, Efault);
1619 unsafe_put_user(info.cause, &infop->si_code, Efault);
1620 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1621 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1622 unsafe_put_user(info.status, &infop->si_status, Efault);
1623 user_access_end();
1624 return err;
1625Efault:
1626 user_access_end();
1627 return -EFAULT;
1628}
1629
1630long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1631 struct rusage *ru)
1632{
1633 struct wait_opts wo;
1634 struct pid *pid = NULL;
1635 enum pid_type type;
1636 long ret;
1637
1638 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1639 __WNOTHREAD|__WCLONE|__WALL))
1640 return -EINVAL;
1641
1642 /* -INT_MIN is not defined */
1643 if (upid == INT_MIN)
1644 return -ESRCH;
1645
1646 if (upid == -1)
1647 type = PIDTYPE_MAX;
1648 else if (upid < 0) {
1649 type = PIDTYPE_PGID;
1650 pid = find_get_pid(-upid);
1651 } else if (upid == 0) {
1652 type = PIDTYPE_PGID;
1653 pid = get_task_pid(current, PIDTYPE_PGID);
1654 } else /* upid > 0 */ {
1655 type = PIDTYPE_PID;
1656 pid = find_get_pid(upid);
1657 }
1658
1659 wo.wo_type = type;
1660 wo.wo_pid = pid;
1661 wo.wo_flags = options | WEXITED;
1662 wo.wo_info = NULL;
1663 wo.wo_stat = 0;
1664 wo.wo_rusage = ru;
1665 ret = do_wait(&wo);
1666 put_pid(pid);
1667 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1668 ret = -EFAULT;
1669
1670 return ret;
1671}
1672
1673SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1674 int, options, struct rusage __user *, ru)
1675{
1676 struct rusage r;
1677 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1678
1679 if (err > 0) {
1680 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1681 return -EFAULT;
1682 }
1683 return err;
1684}
1685
1686#ifdef __ARCH_WANT_SYS_WAITPID
1687
1688/*
1689 * sys_waitpid() remains for compatibility. waitpid() should be
1690 * implemented by calling sys_wait4() from libc.a.
1691 */
1692SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1693{
1694 return kernel_wait4(pid, stat_addr, options, NULL);
1695}
1696
1697#endif
1698
1699#ifdef CONFIG_COMPAT
1700COMPAT_SYSCALL_DEFINE4(wait4,
1701 compat_pid_t, pid,
1702 compat_uint_t __user *, stat_addr,
1703 int, options,
1704 struct compat_rusage __user *, ru)
1705{
1706 struct rusage r;
1707 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1708 if (err > 0) {
1709 if (ru && put_compat_rusage(&r, ru))
1710 return -EFAULT;
1711 }
1712 return err;
1713}
1714
1715COMPAT_SYSCALL_DEFINE5(waitid,
1716 int, which, compat_pid_t, pid,
1717 struct compat_siginfo __user *, infop, int, options,
1718 struct compat_rusage __user *, uru)
1719{
1720 struct rusage ru;
1721 struct waitid_info info = {.status = 0};
1722 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1723 int signo = 0;
1724 if (err > 0) {
1725 signo = SIGCHLD;
1726 err = 0;
1727 if (uru) {
1728 /* kernel_waitid() overwrites everything in ru */
1729 if (COMPAT_USE_64BIT_TIME)
1730 err = copy_to_user(uru, &ru, sizeof(ru));
1731 else
1732 err = put_compat_rusage(&ru, uru);
1733 if (err)
1734 return -EFAULT;
1735 }
1736 }
1737
1738 if (!infop)
1739 return err;
1740
1741 if (!access_ok(VERIFY_WRITE, infop, sizeof(*infop)))
1742 return -EFAULT;
1743
1744 user_access_begin();
1745 unsafe_put_user(signo, &infop->si_signo, Efault);
1746 unsafe_put_user(0, &infop->si_errno, Efault);
1747 unsafe_put_user(info.cause, &infop->si_code, Efault);
1748 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1749 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1750 unsafe_put_user(info.status, &infop->si_status, Efault);
1751 user_access_end();
1752 return err;
1753Efault:
1754 user_access_end();
1755 return -EFAULT;
1756}
1757#endif
1758
1759__weak void abort(void)
1760{
1761 BUG();
1762
1763 /* if that doesn't kill us, halt */
1764 panic("Oops failed to kill thread");
1765}
1766EXPORT_SYMBOL(abort);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/kernel/exit.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8#include <linux/mm.h>
9#include <linux/slab.h>
10#include <linux/sched/autogroup.h>
11#include <linux/sched/mm.h>
12#include <linux/sched/stat.h>
13#include <linux/sched/task.h>
14#include <linux/sched/task_stack.h>
15#include <linux/sched/cputime.h>
16#include <linux/interrupt.h>
17#include <linux/module.h>
18#include <linux/capability.h>
19#include <linux/completion.h>
20#include <linux/personality.h>
21#include <linux/tty.h>
22#include <linux/iocontext.h>
23#include <linux/key.h>
24#include <linux/cpu.h>
25#include <linux/acct.h>
26#include <linux/tsacct_kern.h>
27#include <linux/file.h>
28#include <linux/fdtable.h>
29#include <linux/freezer.h>
30#include <linux/binfmts.h>
31#include <linux/nsproxy.h>
32#include <linux/pid_namespace.h>
33#include <linux/ptrace.h>
34#include <linux/profile.h>
35#include <linux/mount.h>
36#include <linux/proc_fs.h>
37#include <linux/kthread.h>
38#include <linux/mempolicy.h>
39#include <linux/taskstats_kern.h>
40#include <linux/delayacct.h>
41#include <linux/cgroup.h>
42#include <linux/syscalls.h>
43#include <linux/signal.h>
44#include <linux/posix-timers.h>
45#include <linux/cn_proc.h>
46#include <linux/mutex.h>
47#include <linux/futex.h>
48#include <linux/pipe_fs_i.h>
49#include <linux/audit.h> /* for audit_free() */
50#include <linux/resource.h>
51#include <linux/blkdev.h>
52#include <linux/task_io_accounting_ops.h>
53#include <linux/tracehook.h>
54#include <linux/fs_struct.h>
55#include <linux/init_task.h>
56#include <linux/perf_event.h>
57#include <trace/events/sched.h>
58#include <linux/hw_breakpoint.h>
59#include <linux/oom.h>
60#include <linux/writeback.h>
61#include <linux/shm.h>
62#include <linux/kcov.h>
63#include <linux/random.h>
64#include <linux/rcuwait.h>
65#include <linux/compat.h>
66
67#include <linux/uaccess.h>
68#include <asm/unistd.h>
69#include <asm/mmu_context.h>
70
71static void __unhash_process(struct task_struct *p, bool group_dead)
72{
73 nr_threads--;
74 detach_pid(p, PIDTYPE_PID);
75 if (group_dead) {
76 detach_pid(p, PIDTYPE_TGID);
77 detach_pid(p, PIDTYPE_PGID);
78 detach_pid(p, PIDTYPE_SID);
79
80 list_del_rcu(&p->tasks);
81 list_del_init(&p->sibling);
82 __this_cpu_dec(process_counts);
83 }
84 list_del_rcu(&p->thread_group);
85 list_del_rcu(&p->thread_node);
86}
87
88/*
89 * This function expects the tasklist_lock write-locked.
90 */
91static void __exit_signal(struct task_struct *tsk)
92{
93 struct signal_struct *sig = tsk->signal;
94 bool group_dead = thread_group_leader(tsk);
95 struct sighand_struct *sighand;
96 struct tty_struct *tty;
97 u64 utime, stime;
98
99 sighand = rcu_dereference_check(tsk->sighand,
100 lockdep_tasklist_lock_is_held());
101 spin_lock(&sighand->siglock);
102
103#ifdef CONFIG_POSIX_TIMERS
104 posix_cpu_timers_exit(tsk);
105 if (group_dead)
106 posix_cpu_timers_exit_group(tsk);
107#endif
108
109 if (group_dead) {
110 tty = sig->tty;
111 sig->tty = NULL;
112 } else {
113 /*
114 * If there is any task waiting for the group exit
115 * then notify it:
116 */
117 if (sig->notify_count > 0 && !--sig->notify_count)
118 wake_up_process(sig->group_exit_task);
119
120 if (tsk == sig->curr_target)
121 sig->curr_target = next_thread(tsk);
122 }
123
124 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
125 sizeof(unsigned long long));
126
127 /*
128 * Accumulate here the counters for all threads as they die. We could
129 * skip the group leader because it is the last user of signal_struct,
130 * but we want to avoid the race with thread_group_cputime() which can
131 * see the empty ->thread_head list.
132 */
133 task_cputime(tsk, &utime, &stime);
134 write_seqlock(&sig->stats_lock);
135 sig->utime += utime;
136 sig->stime += stime;
137 sig->gtime += task_gtime(tsk);
138 sig->min_flt += tsk->min_flt;
139 sig->maj_flt += tsk->maj_flt;
140 sig->nvcsw += tsk->nvcsw;
141 sig->nivcsw += tsk->nivcsw;
142 sig->inblock += task_io_get_inblock(tsk);
143 sig->oublock += task_io_get_oublock(tsk);
144 task_io_accounting_add(&sig->ioac, &tsk->ioac);
145 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
146 sig->nr_threads--;
147 __unhash_process(tsk, group_dead);
148 write_sequnlock(&sig->stats_lock);
149
150 /*
151 * Do this under ->siglock, we can race with another thread
152 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
153 */
154 flush_sigqueue(&tsk->pending);
155 tsk->sighand = NULL;
156 spin_unlock(&sighand->siglock);
157
158 __cleanup_sighand(sighand);
159 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
160 if (group_dead) {
161 flush_sigqueue(&sig->shared_pending);
162 tty_kref_put(tty);
163 }
164}
165
166static void delayed_put_task_struct(struct rcu_head *rhp)
167{
168 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
169
170 perf_event_delayed_put(tsk);
171 trace_sched_process_free(tsk);
172 put_task_struct(tsk);
173}
174
175void put_task_struct_rcu_user(struct task_struct *task)
176{
177 if (refcount_dec_and_test(&task->rcu_users))
178 call_rcu(&task->rcu, delayed_put_task_struct);
179}
180
181void release_task(struct task_struct *p)
182{
183 struct task_struct *leader;
184 struct pid *thread_pid;
185 int zap_leader;
186repeat:
187 /* don't need to get the RCU readlock here - the process is dead and
188 * can't be modifying its own credentials. But shut RCU-lockdep up */
189 rcu_read_lock();
190 atomic_dec(&__task_cred(p)->user->processes);
191 rcu_read_unlock();
192
193 cgroup_release(p);
194
195 write_lock_irq(&tasklist_lock);
196 ptrace_release_task(p);
197 thread_pid = get_pid(p->thread_pid);
198 __exit_signal(p);
199
200 /*
201 * If we are the last non-leader member of the thread
202 * group, and the leader is zombie, then notify the
203 * group leader's parent process. (if it wants notification.)
204 */
205 zap_leader = 0;
206 leader = p->group_leader;
207 if (leader != p && thread_group_empty(leader)
208 && leader->exit_state == EXIT_ZOMBIE) {
209 /*
210 * If we were the last child thread and the leader has
211 * exited already, and the leader's parent ignores SIGCHLD,
212 * then we are the one who should release the leader.
213 */
214 zap_leader = do_notify_parent(leader, leader->exit_signal);
215 if (zap_leader)
216 leader->exit_state = EXIT_DEAD;
217 }
218
219 write_unlock_irq(&tasklist_lock);
220 seccomp_filter_release(p);
221 proc_flush_pid(thread_pid);
222 put_pid(thread_pid);
223 release_thread(p);
224 put_task_struct_rcu_user(p);
225
226 p = leader;
227 if (unlikely(zap_leader))
228 goto repeat;
229}
230
231int rcuwait_wake_up(struct rcuwait *w)
232{
233 int ret = 0;
234 struct task_struct *task;
235
236 rcu_read_lock();
237
238 /*
239 * Order condition vs @task, such that everything prior to the load
240 * of @task is visible. This is the condition as to why the user called
241 * rcuwait_wake() in the first place. Pairs with set_current_state()
242 * barrier (A) in rcuwait_wait_event().
243 *
244 * WAIT WAKE
245 * [S] tsk = current [S] cond = true
246 * MB (A) MB (B)
247 * [L] cond [L] tsk
248 */
249 smp_mb(); /* (B) */
250
251 task = rcu_dereference(w->task);
252 if (task)
253 ret = wake_up_process(task);
254 rcu_read_unlock();
255
256 return ret;
257}
258EXPORT_SYMBOL_GPL(rcuwait_wake_up);
259
260/*
261 * Determine if a process group is "orphaned", according to the POSIX
262 * definition in 2.2.2.52. Orphaned process groups are not to be affected
263 * by terminal-generated stop signals. Newly orphaned process groups are
264 * to receive a SIGHUP and a SIGCONT.
265 *
266 * "I ask you, have you ever known what it is to be an orphan?"
267 */
268static int will_become_orphaned_pgrp(struct pid *pgrp,
269 struct task_struct *ignored_task)
270{
271 struct task_struct *p;
272
273 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
274 if ((p == ignored_task) ||
275 (p->exit_state && thread_group_empty(p)) ||
276 is_global_init(p->real_parent))
277 continue;
278
279 if (task_pgrp(p->real_parent) != pgrp &&
280 task_session(p->real_parent) == task_session(p))
281 return 0;
282 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
283
284 return 1;
285}
286
287int is_current_pgrp_orphaned(void)
288{
289 int retval;
290
291 read_lock(&tasklist_lock);
292 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
293 read_unlock(&tasklist_lock);
294
295 return retval;
296}
297
298static bool has_stopped_jobs(struct pid *pgrp)
299{
300 struct task_struct *p;
301
302 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
303 if (p->signal->flags & SIGNAL_STOP_STOPPED)
304 return true;
305 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
306
307 return false;
308}
309
310/*
311 * Check to see if any process groups have become orphaned as
312 * a result of our exiting, and if they have any stopped jobs,
313 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
314 */
315static void
316kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
317{
318 struct pid *pgrp = task_pgrp(tsk);
319 struct task_struct *ignored_task = tsk;
320
321 if (!parent)
322 /* exit: our father is in a different pgrp than
323 * we are and we were the only connection outside.
324 */
325 parent = tsk->real_parent;
326 else
327 /* reparent: our child is in a different pgrp than
328 * we are, and it was the only connection outside.
329 */
330 ignored_task = NULL;
331
332 if (task_pgrp(parent) != pgrp &&
333 task_session(parent) == task_session(tsk) &&
334 will_become_orphaned_pgrp(pgrp, ignored_task) &&
335 has_stopped_jobs(pgrp)) {
336 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
337 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
338 }
339}
340
341#ifdef CONFIG_MEMCG
342/*
343 * A task is exiting. If it owned this mm, find a new owner for the mm.
344 */
345void mm_update_next_owner(struct mm_struct *mm)
346{
347 struct task_struct *c, *g, *p = current;
348
349retry:
350 /*
351 * If the exiting or execing task is not the owner, it's
352 * someone else's problem.
353 */
354 if (mm->owner != p)
355 return;
356 /*
357 * The current owner is exiting/execing and there are no other
358 * candidates. Do not leave the mm pointing to a possibly
359 * freed task structure.
360 */
361 if (atomic_read(&mm->mm_users) <= 1) {
362 WRITE_ONCE(mm->owner, NULL);
363 return;
364 }
365
366 read_lock(&tasklist_lock);
367 /*
368 * Search in the children
369 */
370 list_for_each_entry(c, &p->children, sibling) {
371 if (c->mm == mm)
372 goto assign_new_owner;
373 }
374
375 /*
376 * Search in the siblings
377 */
378 list_for_each_entry(c, &p->real_parent->children, sibling) {
379 if (c->mm == mm)
380 goto assign_new_owner;
381 }
382
383 /*
384 * Search through everything else, we should not get here often.
385 */
386 for_each_process(g) {
387 if (g->flags & PF_KTHREAD)
388 continue;
389 for_each_thread(g, c) {
390 if (c->mm == mm)
391 goto assign_new_owner;
392 if (c->mm)
393 break;
394 }
395 }
396 read_unlock(&tasklist_lock);
397 /*
398 * We found no owner yet mm_users > 1: this implies that we are
399 * most likely racing with swapoff (try_to_unuse()) or /proc or
400 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
401 */
402 WRITE_ONCE(mm->owner, NULL);
403 return;
404
405assign_new_owner:
406 BUG_ON(c == p);
407 get_task_struct(c);
408 /*
409 * The task_lock protects c->mm from changing.
410 * We always want mm->owner->mm == mm
411 */
412 task_lock(c);
413 /*
414 * Delay read_unlock() till we have the task_lock()
415 * to ensure that c does not slip away underneath us
416 */
417 read_unlock(&tasklist_lock);
418 if (c->mm != mm) {
419 task_unlock(c);
420 put_task_struct(c);
421 goto retry;
422 }
423 WRITE_ONCE(mm->owner, c);
424 task_unlock(c);
425 put_task_struct(c);
426}
427#endif /* CONFIG_MEMCG */
428
429/*
430 * Turn us into a lazy TLB process if we
431 * aren't already..
432 */
433static void exit_mm(void)
434{
435 struct mm_struct *mm = current->mm;
436 struct core_state *core_state;
437
438 exit_mm_release(current, mm);
439 if (!mm)
440 return;
441 sync_mm_rss(mm);
442 /*
443 * Serialize with any possible pending coredump.
444 * We must hold mmap_lock around checking core_state
445 * and clearing tsk->mm. The core-inducing thread
446 * will increment ->nr_threads for each thread in the
447 * group with ->mm != NULL.
448 */
449 mmap_read_lock(mm);
450 core_state = mm->core_state;
451 if (core_state) {
452 struct core_thread self;
453
454 mmap_read_unlock(mm);
455
456 self.task = current;
457 self.next = xchg(&core_state->dumper.next, &self);
458 /*
459 * Implies mb(), the result of xchg() must be visible
460 * to core_state->dumper.
461 */
462 if (atomic_dec_and_test(&core_state->nr_threads))
463 complete(&core_state->startup);
464
465 for (;;) {
466 set_current_state(TASK_UNINTERRUPTIBLE);
467 if (!self.task) /* see coredump_finish() */
468 break;
469 freezable_schedule();
470 }
471 __set_current_state(TASK_RUNNING);
472 mmap_read_lock(mm);
473 }
474 mmgrab(mm);
475 BUG_ON(mm != current->active_mm);
476 /* more a memory barrier than a real lock */
477 task_lock(current);
478 current->mm = NULL;
479 mmap_read_unlock(mm);
480 enter_lazy_tlb(mm, current);
481 task_unlock(current);
482 mm_update_next_owner(mm);
483 mmput(mm);
484 if (test_thread_flag(TIF_MEMDIE))
485 exit_oom_victim();
486}
487
488static struct task_struct *find_alive_thread(struct task_struct *p)
489{
490 struct task_struct *t;
491
492 for_each_thread(p, t) {
493 if (!(t->flags & PF_EXITING))
494 return t;
495 }
496 return NULL;
497}
498
499static struct task_struct *find_child_reaper(struct task_struct *father,
500 struct list_head *dead)
501 __releases(&tasklist_lock)
502 __acquires(&tasklist_lock)
503{
504 struct pid_namespace *pid_ns = task_active_pid_ns(father);
505 struct task_struct *reaper = pid_ns->child_reaper;
506 struct task_struct *p, *n;
507
508 if (likely(reaper != father))
509 return reaper;
510
511 reaper = find_alive_thread(father);
512 if (reaper) {
513 pid_ns->child_reaper = reaper;
514 return reaper;
515 }
516
517 write_unlock_irq(&tasklist_lock);
518
519 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
520 list_del_init(&p->ptrace_entry);
521 release_task(p);
522 }
523
524 zap_pid_ns_processes(pid_ns);
525 write_lock_irq(&tasklist_lock);
526
527 return father;
528}
529
530/*
531 * When we die, we re-parent all our children, and try to:
532 * 1. give them to another thread in our thread group, if such a member exists
533 * 2. give it to the first ancestor process which prctl'd itself as a
534 * child_subreaper for its children (like a service manager)
535 * 3. give it to the init process (PID 1) in our pid namespace
536 */
537static struct task_struct *find_new_reaper(struct task_struct *father,
538 struct task_struct *child_reaper)
539{
540 struct task_struct *thread, *reaper;
541
542 thread = find_alive_thread(father);
543 if (thread)
544 return thread;
545
546 if (father->signal->has_child_subreaper) {
547 unsigned int ns_level = task_pid(father)->level;
548 /*
549 * Find the first ->is_child_subreaper ancestor in our pid_ns.
550 * We can't check reaper != child_reaper to ensure we do not
551 * cross the namespaces, the exiting parent could be injected
552 * by setns() + fork().
553 * We check pid->level, this is slightly more efficient than
554 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
555 */
556 for (reaper = father->real_parent;
557 task_pid(reaper)->level == ns_level;
558 reaper = reaper->real_parent) {
559 if (reaper == &init_task)
560 break;
561 if (!reaper->signal->is_child_subreaper)
562 continue;
563 thread = find_alive_thread(reaper);
564 if (thread)
565 return thread;
566 }
567 }
568
569 return child_reaper;
570}
571
572/*
573* Any that need to be release_task'd are put on the @dead list.
574 */
575static void reparent_leader(struct task_struct *father, struct task_struct *p,
576 struct list_head *dead)
577{
578 if (unlikely(p->exit_state == EXIT_DEAD))
579 return;
580
581 /* We don't want people slaying init. */
582 p->exit_signal = SIGCHLD;
583
584 /* If it has exited notify the new parent about this child's death. */
585 if (!p->ptrace &&
586 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
587 if (do_notify_parent(p, p->exit_signal)) {
588 p->exit_state = EXIT_DEAD;
589 list_add(&p->ptrace_entry, dead);
590 }
591 }
592
593 kill_orphaned_pgrp(p, father);
594}
595
596/*
597 * This does two things:
598 *
599 * A. Make init inherit all the child processes
600 * B. Check to see if any process groups have become orphaned
601 * as a result of our exiting, and if they have any stopped
602 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
603 */
604static void forget_original_parent(struct task_struct *father,
605 struct list_head *dead)
606{
607 struct task_struct *p, *t, *reaper;
608
609 if (unlikely(!list_empty(&father->ptraced)))
610 exit_ptrace(father, dead);
611
612 /* Can drop and reacquire tasklist_lock */
613 reaper = find_child_reaper(father, dead);
614 if (list_empty(&father->children))
615 return;
616
617 reaper = find_new_reaper(father, reaper);
618 list_for_each_entry(p, &father->children, sibling) {
619 for_each_thread(p, t) {
620 RCU_INIT_POINTER(t->real_parent, reaper);
621 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
622 if (likely(!t->ptrace))
623 t->parent = t->real_parent;
624 if (t->pdeath_signal)
625 group_send_sig_info(t->pdeath_signal,
626 SEND_SIG_NOINFO, t,
627 PIDTYPE_TGID);
628 }
629 /*
630 * If this is a threaded reparent there is no need to
631 * notify anyone anything has happened.
632 */
633 if (!same_thread_group(reaper, father))
634 reparent_leader(father, p, dead);
635 }
636 list_splice_tail_init(&father->children, &reaper->children);
637}
638
639/*
640 * Send signals to all our closest relatives so that they know
641 * to properly mourn us..
642 */
643static void exit_notify(struct task_struct *tsk, int group_dead)
644{
645 bool autoreap;
646 struct task_struct *p, *n;
647 LIST_HEAD(dead);
648
649 write_lock_irq(&tasklist_lock);
650 forget_original_parent(tsk, &dead);
651
652 if (group_dead)
653 kill_orphaned_pgrp(tsk->group_leader, NULL);
654
655 tsk->exit_state = EXIT_ZOMBIE;
656 if (unlikely(tsk->ptrace)) {
657 int sig = thread_group_leader(tsk) &&
658 thread_group_empty(tsk) &&
659 !ptrace_reparented(tsk) ?
660 tsk->exit_signal : SIGCHLD;
661 autoreap = do_notify_parent(tsk, sig);
662 } else if (thread_group_leader(tsk)) {
663 autoreap = thread_group_empty(tsk) &&
664 do_notify_parent(tsk, tsk->exit_signal);
665 } else {
666 autoreap = true;
667 }
668
669 if (autoreap) {
670 tsk->exit_state = EXIT_DEAD;
671 list_add(&tsk->ptrace_entry, &dead);
672 }
673
674 /* mt-exec, de_thread() is waiting for group leader */
675 if (unlikely(tsk->signal->notify_count < 0))
676 wake_up_process(tsk->signal->group_exit_task);
677 write_unlock_irq(&tasklist_lock);
678
679 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
680 list_del_init(&p->ptrace_entry);
681 release_task(p);
682 }
683}
684
685#ifdef CONFIG_DEBUG_STACK_USAGE
686static void check_stack_usage(void)
687{
688 static DEFINE_SPINLOCK(low_water_lock);
689 static int lowest_to_date = THREAD_SIZE;
690 unsigned long free;
691
692 free = stack_not_used(current);
693
694 if (free >= lowest_to_date)
695 return;
696
697 spin_lock(&low_water_lock);
698 if (free < lowest_to_date) {
699 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
700 current->comm, task_pid_nr(current), free);
701 lowest_to_date = free;
702 }
703 spin_unlock(&low_water_lock);
704}
705#else
706static inline void check_stack_usage(void) {}
707#endif
708
709void __noreturn do_exit(long code)
710{
711 struct task_struct *tsk = current;
712 int group_dead;
713
714 /*
715 * We can get here from a kernel oops, sometimes with preemption off.
716 * Start by checking for critical errors.
717 * Then fix up important state like USER_DS and preemption.
718 * Then do everything else.
719 */
720
721 WARN_ON(blk_needs_flush_plug(tsk));
722
723 if (unlikely(in_interrupt()))
724 panic("Aiee, killing interrupt handler!");
725 if (unlikely(!tsk->pid))
726 panic("Attempted to kill the idle task!");
727
728 /*
729 * If do_exit is called because this processes oopsed, it's possible
730 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
731 * continuing. Amongst other possible reasons, this is to prevent
732 * mm_release()->clear_child_tid() from writing to a user-controlled
733 * kernel address.
734 */
735 force_uaccess_begin();
736
737 if (unlikely(in_atomic())) {
738 pr_info("note: %s[%d] exited with preempt_count %d\n",
739 current->comm, task_pid_nr(current),
740 preempt_count());
741 preempt_count_set(PREEMPT_ENABLED);
742 }
743
744 profile_task_exit(tsk);
745 kcov_task_exit(tsk);
746
747 ptrace_event(PTRACE_EVENT_EXIT, code);
748
749 validate_creds_for_do_exit(tsk);
750
751 /*
752 * We're taking recursive faults here in do_exit. Safest is to just
753 * leave this task alone and wait for reboot.
754 */
755 if (unlikely(tsk->flags & PF_EXITING)) {
756 pr_alert("Fixing recursive fault but reboot is needed!\n");
757 futex_exit_recursive(tsk);
758 set_current_state(TASK_UNINTERRUPTIBLE);
759 schedule();
760 }
761
762 exit_signals(tsk); /* sets PF_EXITING */
763
764 /* sync mm's RSS info before statistics gathering */
765 if (tsk->mm)
766 sync_mm_rss(tsk->mm);
767 acct_update_integrals(tsk);
768 group_dead = atomic_dec_and_test(&tsk->signal->live);
769 if (group_dead) {
770 /*
771 * If the last thread of global init has exited, panic
772 * immediately to get a useable coredump.
773 */
774 if (unlikely(is_global_init(tsk)))
775 panic("Attempted to kill init! exitcode=0x%08x\n",
776 tsk->signal->group_exit_code ?: (int)code);
777
778#ifdef CONFIG_POSIX_TIMERS
779 hrtimer_cancel(&tsk->signal->real_timer);
780 exit_itimers(tsk->signal);
781#endif
782 if (tsk->mm)
783 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
784 }
785 acct_collect(code, group_dead);
786 if (group_dead)
787 tty_audit_exit();
788 audit_free(tsk);
789
790 tsk->exit_code = code;
791 taskstats_exit(tsk, group_dead);
792
793 exit_mm();
794
795 if (group_dead)
796 acct_process();
797 trace_sched_process_exit(tsk);
798
799 exit_sem(tsk);
800 exit_shm(tsk);
801 exit_files(tsk);
802 exit_fs(tsk);
803 if (group_dead)
804 disassociate_ctty(1);
805 exit_task_namespaces(tsk);
806 exit_task_work(tsk);
807 exit_thread(tsk);
808
809 /*
810 * Flush inherited counters to the parent - before the parent
811 * gets woken up by child-exit notifications.
812 *
813 * because of cgroup mode, must be called before cgroup_exit()
814 */
815 perf_event_exit_task(tsk);
816
817 sched_autogroup_exit_task(tsk);
818 cgroup_exit(tsk);
819
820 /*
821 * FIXME: do that only when needed, using sched_exit tracepoint
822 */
823 flush_ptrace_hw_breakpoint(tsk);
824
825 exit_tasks_rcu_start();
826 exit_notify(tsk, group_dead);
827 proc_exit_connector(tsk);
828 mpol_put_task_policy(tsk);
829#ifdef CONFIG_FUTEX
830 if (unlikely(current->pi_state_cache))
831 kfree(current->pi_state_cache);
832#endif
833 /*
834 * Make sure we are holding no locks:
835 */
836 debug_check_no_locks_held();
837
838 if (tsk->io_context)
839 exit_io_context(tsk);
840
841 if (tsk->splice_pipe)
842 free_pipe_info(tsk->splice_pipe);
843
844 if (tsk->task_frag.page)
845 put_page(tsk->task_frag.page);
846
847 validate_creds_for_do_exit(tsk);
848
849 check_stack_usage();
850 preempt_disable();
851 if (tsk->nr_dirtied)
852 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
853 exit_rcu();
854 exit_tasks_rcu_finish();
855
856 lockdep_free_task(tsk);
857 do_task_dead();
858}
859EXPORT_SYMBOL_GPL(do_exit);
860
861void complete_and_exit(struct completion *comp, long code)
862{
863 if (comp)
864 complete(comp);
865
866 do_exit(code);
867}
868EXPORT_SYMBOL(complete_and_exit);
869
870SYSCALL_DEFINE1(exit, int, error_code)
871{
872 do_exit((error_code&0xff)<<8);
873}
874
875/*
876 * Take down every thread in the group. This is called by fatal signals
877 * as well as by sys_exit_group (below).
878 */
879void
880do_group_exit(int exit_code)
881{
882 struct signal_struct *sig = current->signal;
883
884 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
885
886 if (signal_group_exit(sig))
887 exit_code = sig->group_exit_code;
888 else if (!thread_group_empty(current)) {
889 struct sighand_struct *const sighand = current->sighand;
890
891 spin_lock_irq(&sighand->siglock);
892 if (signal_group_exit(sig))
893 /* Another thread got here before we took the lock. */
894 exit_code = sig->group_exit_code;
895 else {
896 sig->group_exit_code = exit_code;
897 sig->flags = SIGNAL_GROUP_EXIT;
898 zap_other_threads(current);
899 }
900 spin_unlock_irq(&sighand->siglock);
901 }
902
903 do_exit(exit_code);
904 /* NOTREACHED */
905}
906
907/*
908 * this kills every thread in the thread group. Note that any externally
909 * wait4()-ing process will get the correct exit code - even if this
910 * thread is not the thread group leader.
911 */
912SYSCALL_DEFINE1(exit_group, int, error_code)
913{
914 do_group_exit((error_code & 0xff) << 8);
915 /* NOTREACHED */
916 return 0;
917}
918
919struct waitid_info {
920 pid_t pid;
921 uid_t uid;
922 int status;
923 int cause;
924};
925
926struct wait_opts {
927 enum pid_type wo_type;
928 int wo_flags;
929 struct pid *wo_pid;
930
931 struct waitid_info *wo_info;
932 int wo_stat;
933 struct rusage *wo_rusage;
934
935 wait_queue_entry_t child_wait;
936 int notask_error;
937};
938
939static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
940{
941 return wo->wo_type == PIDTYPE_MAX ||
942 task_pid_type(p, wo->wo_type) == wo->wo_pid;
943}
944
945static int
946eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
947{
948 if (!eligible_pid(wo, p))
949 return 0;
950
951 /*
952 * Wait for all children (clone and not) if __WALL is set or
953 * if it is traced by us.
954 */
955 if (ptrace || (wo->wo_flags & __WALL))
956 return 1;
957
958 /*
959 * Otherwise, wait for clone children *only* if __WCLONE is set;
960 * otherwise, wait for non-clone children *only*.
961 *
962 * Note: a "clone" child here is one that reports to its parent
963 * using a signal other than SIGCHLD, or a non-leader thread which
964 * we can only see if it is traced by us.
965 */
966 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
967 return 0;
968
969 return 1;
970}
971
972/*
973 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
974 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
975 * the lock and this task is uninteresting. If we return nonzero, we have
976 * released the lock and the system call should return.
977 */
978static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
979{
980 int state, status;
981 pid_t pid = task_pid_vnr(p);
982 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
983 struct waitid_info *infop;
984
985 if (!likely(wo->wo_flags & WEXITED))
986 return 0;
987
988 if (unlikely(wo->wo_flags & WNOWAIT)) {
989 status = p->exit_code;
990 get_task_struct(p);
991 read_unlock(&tasklist_lock);
992 sched_annotate_sleep();
993 if (wo->wo_rusage)
994 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
995 put_task_struct(p);
996 goto out_info;
997 }
998 /*
999 * Move the task's state to DEAD/TRACE, only one thread can do this.
1000 */
1001 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1002 EXIT_TRACE : EXIT_DEAD;
1003 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1004 return 0;
1005 /*
1006 * We own this thread, nobody else can reap it.
1007 */
1008 read_unlock(&tasklist_lock);
1009 sched_annotate_sleep();
1010
1011 /*
1012 * Check thread_group_leader() to exclude the traced sub-threads.
1013 */
1014 if (state == EXIT_DEAD && thread_group_leader(p)) {
1015 struct signal_struct *sig = p->signal;
1016 struct signal_struct *psig = current->signal;
1017 unsigned long maxrss;
1018 u64 tgutime, tgstime;
1019
1020 /*
1021 * The resource counters for the group leader are in its
1022 * own task_struct. Those for dead threads in the group
1023 * are in its signal_struct, as are those for the child
1024 * processes it has previously reaped. All these
1025 * accumulate in the parent's signal_struct c* fields.
1026 *
1027 * We don't bother to take a lock here to protect these
1028 * p->signal fields because the whole thread group is dead
1029 * and nobody can change them.
1030 *
1031 * psig->stats_lock also protects us from our sub-theads
1032 * which can reap other children at the same time. Until
1033 * we change k_getrusage()-like users to rely on this lock
1034 * we have to take ->siglock as well.
1035 *
1036 * We use thread_group_cputime_adjusted() to get times for
1037 * the thread group, which consolidates times for all threads
1038 * in the group including the group leader.
1039 */
1040 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1041 spin_lock_irq(¤t->sighand->siglock);
1042 write_seqlock(&psig->stats_lock);
1043 psig->cutime += tgutime + sig->cutime;
1044 psig->cstime += tgstime + sig->cstime;
1045 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1046 psig->cmin_flt +=
1047 p->min_flt + sig->min_flt + sig->cmin_flt;
1048 psig->cmaj_flt +=
1049 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1050 psig->cnvcsw +=
1051 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1052 psig->cnivcsw +=
1053 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1054 psig->cinblock +=
1055 task_io_get_inblock(p) +
1056 sig->inblock + sig->cinblock;
1057 psig->coublock +=
1058 task_io_get_oublock(p) +
1059 sig->oublock + sig->coublock;
1060 maxrss = max(sig->maxrss, sig->cmaxrss);
1061 if (psig->cmaxrss < maxrss)
1062 psig->cmaxrss = maxrss;
1063 task_io_accounting_add(&psig->ioac, &p->ioac);
1064 task_io_accounting_add(&psig->ioac, &sig->ioac);
1065 write_sequnlock(&psig->stats_lock);
1066 spin_unlock_irq(¤t->sighand->siglock);
1067 }
1068
1069 if (wo->wo_rusage)
1070 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1071 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1072 ? p->signal->group_exit_code : p->exit_code;
1073 wo->wo_stat = status;
1074
1075 if (state == EXIT_TRACE) {
1076 write_lock_irq(&tasklist_lock);
1077 /* We dropped tasklist, ptracer could die and untrace */
1078 ptrace_unlink(p);
1079
1080 /* If parent wants a zombie, don't release it now */
1081 state = EXIT_ZOMBIE;
1082 if (do_notify_parent(p, p->exit_signal))
1083 state = EXIT_DEAD;
1084 p->exit_state = state;
1085 write_unlock_irq(&tasklist_lock);
1086 }
1087 if (state == EXIT_DEAD)
1088 release_task(p);
1089
1090out_info:
1091 infop = wo->wo_info;
1092 if (infop) {
1093 if ((status & 0x7f) == 0) {
1094 infop->cause = CLD_EXITED;
1095 infop->status = status >> 8;
1096 } else {
1097 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1098 infop->status = status & 0x7f;
1099 }
1100 infop->pid = pid;
1101 infop->uid = uid;
1102 }
1103
1104 return pid;
1105}
1106
1107static int *task_stopped_code(struct task_struct *p, bool ptrace)
1108{
1109 if (ptrace) {
1110 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1111 return &p->exit_code;
1112 } else {
1113 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1114 return &p->signal->group_exit_code;
1115 }
1116 return NULL;
1117}
1118
1119/**
1120 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1121 * @wo: wait options
1122 * @ptrace: is the wait for ptrace
1123 * @p: task to wait for
1124 *
1125 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1126 *
1127 * CONTEXT:
1128 * read_lock(&tasklist_lock), which is released if return value is
1129 * non-zero. Also, grabs and releases @p->sighand->siglock.
1130 *
1131 * RETURNS:
1132 * 0 if wait condition didn't exist and search for other wait conditions
1133 * should continue. Non-zero return, -errno on failure and @p's pid on
1134 * success, implies that tasklist_lock is released and wait condition
1135 * search should terminate.
1136 */
1137static int wait_task_stopped(struct wait_opts *wo,
1138 int ptrace, struct task_struct *p)
1139{
1140 struct waitid_info *infop;
1141 int exit_code, *p_code, why;
1142 uid_t uid = 0; /* unneeded, required by compiler */
1143 pid_t pid;
1144
1145 /*
1146 * Traditionally we see ptrace'd stopped tasks regardless of options.
1147 */
1148 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1149 return 0;
1150
1151 if (!task_stopped_code(p, ptrace))
1152 return 0;
1153
1154 exit_code = 0;
1155 spin_lock_irq(&p->sighand->siglock);
1156
1157 p_code = task_stopped_code(p, ptrace);
1158 if (unlikely(!p_code))
1159 goto unlock_sig;
1160
1161 exit_code = *p_code;
1162 if (!exit_code)
1163 goto unlock_sig;
1164
1165 if (!unlikely(wo->wo_flags & WNOWAIT))
1166 *p_code = 0;
1167
1168 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1169unlock_sig:
1170 spin_unlock_irq(&p->sighand->siglock);
1171 if (!exit_code)
1172 return 0;
1173
1174 /*
1175 * Now we are pretty sure this task is interesting.
1176 * Make sure it doesn't get reaped out from under us while we
1177 * give up the lock and then examine it below. We don't want to
1178 * keep holding onto the tasklist_lock while we call getrusage and
1179 * possibly take page faults for user memory.
1180 */
1181 get_task_struct(p);
1182 pid = task_pid_vnr(p);
1183 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1184 read_unlock(&tasklist_lock);
1185 sched_annotate_sleep();
1186 if (wo->wo_rusage)
1187 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1188 put_task_struct(p);
1189
1190 if (likely(!(wo->wo_flags & WNOWAIT)))
1191 wo->wo_stat = (exit_code << 8) | 0x7f;
1192
1193 infop = wo->wo_info;
1194 if (infop) {
1195 infop->cause = why;
1196 infop->status = exit_code;
1197 infop->pid = pid;
1198 infop->uid = uid;
1199 }
1200 return pid;
1201}
1202
1203/*
1204 * Handle do_wait work for one task in a live, non-stopped state.
1205 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1206 * the lock and this task is uninteresting. If we return nonzero, we have
1207 * released the lock and the system call should return.
1208 */
1209static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1210{
1211 struct waitid_info *infop;
1212 pid_t pid;
1213 uid_t uid;
1214
1215 if (!unlikely(wo->wo_flags & WCONTINUED))
1216 return 0;
1217
1218 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1219 return 0;
1220
1221 spin_lock_irq(&p->sighand->siglock);
1222 /* Re-check with the lock held. */
1223 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1224 spin_unlock_irq(&p->sighand->siglock);
1225 return 0;
1226 }
1227 if (!unlikely(wo->wo_flags & WNOWAIT))
1228 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1229 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1230 spin_unlock_irq(&p->sighand->siglock);
1231
1232 pid = task_pid_vnr(p);
1233 get_task_struct(p);
1234 read_unlock(&tasklist_lock);
1235 sched_annotate_sleep();
1236 if (wo->wo_rusage)
1237 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1238 put_task_struct(p);
1239
1240 infop = wo->wo_info;
1241 if (!infop) {
1242 wo->wo_stat = 0xffff;
1243 } else {
1244 infop->cause = CLD_CONTINUED;
1245 infop->pid = pid;
1246 infop->uid = uid;
1247 infop->status = SIGCONT;
1248 }
1249 return pid;
1250}
1251
1252/*
1253 * Consider @p for a wait by @parent.
1254 *
1255 * -ECHILD should be in ->notask_error before the first call.
1256 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1257 * Returns zero if the search for a child should continue;
1258 * then ->notask_error is 0 if @p is an eligible child,
1259 * or still -ECHILD.
1260 */
1261static int wait_consider_task(struct wait_opts *wo, int ptrace,
1262 struct task_struct *p)
1263{
1264 /*
1265 * We can race with wait_task_zombie() from another thread.
1266 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1267 * can't confuse the checks below.
1268 */
1269 int exit_state = READ_ONCE(p->exit_state);
1270 int ret;
1271
1272 if (unlikely(exit_state == EXIT_DEAD))
1273 return 0;
1274
1275 ret = eligible_child(wo, ptrace, p);
1276 if (!ret)
1277 return ret;
1278
1279 if (unlikely(exit_state == EXIT_TRACE)) {
1280 /*
1281 * ptrace == 0 means we are the natural parent. In this case
1282 * we should clear notask_error, debugger will notify us.
1283 */
1284 if (likely(!ptrace))
1285 wo->notask_error = 0;
1286 return 0;
1287 }
1288
1289 if (likely(!ptrace) && unlikely(p->ptrace)) {
1290 /*
1291 * If it is traced by its real parent's group, just pretend
1292 * the caller is ptrace_do_wait() and reap this child if it
1293 * is zombie.
1294 *
1295 * This also hides group stop state from real parent; otherwise
1296 * a single stop can be reported twice as group and ptrace stop.
1297 * If a ptracer wants to distinguish these two events for its
1298 * own children it should create a separate process which takes
1299 * the role of real parent.
1300 */
1301 if (!ptrace_reparented(p))
1302 ptrace = 1;
1303 }
1304
1305 /* slay zombie? */
1306 if (exit_state == EXIT_ZOMBIE) {
1307 /* we don't reap group leaders with subthreads */
1308 if (!delay_group_leader(p)) {
1309 /*
1310 * A zombie ptracee is only visible to its ptracer.
1311 * Notification and reaping will be cascaded to the
1312 * real parent when the ptracer detaches.
1313 */
1314 if (unlikely(ptrace) || likely(!p->ptrace))
1315 return wait_task_zombie(wo, p);
1316 }
1317
1318 /*
1319 * Allow access to stopped/continued state via zombie by
1320 * falling through. Clearing of notask_error is complex.
1321 *
1322 * When !@ptrace:
1323 *
1324 * If WEXITED is set, notask_error should naturally be
1325 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1326 * so, if there are live subthreads, there are events to
1327 * wait for. If all subthreads are dead, it's still safe
1328 * to clear - this function will be called again in finite
1329 * amount time once all the subthreads are released and
1330 * will then return without clearing.
1331 *
1332 * When @ptrace:
1333 *
1334 * Stopped state is per-task and thus can't change once the
1335 * target task dies. Only continued and exited can happen.
1336 * Clear notask_error if WCONTINUED | WEXITED.
1337 */
1338 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1339 wo->notask_error = 0;
1340 } else {
1341 /*
1342 * @p is alive and it's gonna stop, continue or exit, so
1343 * there always is something to wait for.
1344 */
1345 wo->notask_error = 0;
1346 }
1347
1348 /*
1349 * Wait for stopped. Depending on @ptrace, different stopped state
1350 * is used and the two don't interact with each other.
1351 */
1352 ret = wait_task_stopped(wo, ptrace, p);
1353 if (ret)
1354 return ret;
1355
1356 /*
1357 * Wait for continued. There's only one continued state and the
1358 * ptracer can consume it which can confuse the real parent. Don't
1359 * use WCONTINUED from ptracer. You don't need or want it.
1360 */
1361 return wait_task_continued(wo, p);
1362}
1363
1364/*
1365 * Do the work of do_wait() for one thread in the group, @tsk.
1366 *
1367 * -ECHILD should be in ->notask_error before the first call.
1368 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1369 * Returns zero if the search for a child should continue; then
1370 * ->notask_error is 0 if there were any eligible children,
1371 * or still -ECHILD.
1372 */
1373static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1374{
1375 struct task_struct *p;
1376
1377 list_for_each_entry(p, &tsk->children, sibling) {
1378 int ret = wait_consider_task(wo, 0, p);
1379
1380 if (ret)
1381 return ret;
1382 }
1383
1384 return 0;
1385}
1386
1387static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1388{
1389 struct task_struct *p;
1390
1391 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1392 int ret = wait_consider_task(wo, 1, p);
1393
1394 if (ret)
1395 return ret;
1396 }
1397
1398 return 0;
1399}
1400
1401static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1402 int sync, void *key)
1403{
1404 struct wait_opts *wo = container_of(wait, struct wait_opts,
1405 child_wait);
1406 struct task_struct *p = key;
1407
1408 if (!eligible_pid(wo, p))
1409 return 0;
1410
1411 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1412 return 0;
1413
1414 return default_wake_function(wait, mode, sync, key);
1415}
1416
1417void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1418{
1419 __wake_up_sync_key(&parent->signal->wait_chldexit,
1420 TASK_INTERRUPTIBLE, p);
1421}
1422
1423static long do_wait(struct wait_opts *wo)
1424{
1425 struct task_struct *tsk;
1426 int retval;
1427
1428 trace_sched_process_wait(wo->wo_pid);
1429
1430 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1431 wo->child_wait.private = current;
1432 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1433repeat:
1434 /*
1435 * If there is nothing that can match our criteria, just get out.
1436 * We will clear ->notask_error to zero if we see any child that
1437 * might later match our criteria, even if we are not able to reap
1438 * it yet.
1439 */
1440 wo->notask_error = -ECHILD;
1441 if ((wo->wo_type < PIDTYPE_MAX) &&
1442 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1443 goto notask;
1444
1445 set_current_state(TASK_INTERRUPTIBLE);
1446 read_lock(&tasklist_lock);
1447 tsk = current;
1448 do {
1449 retval = do_wait_thread(wo, tsk);
1450 if (retval)
1451 goto end;
1452
1453 retval = ptrace_do_wait(wo, tsk);
1454 if (retval)
1455 goto end;
1456
1457 if (wo->wo_flags & __WNOTHREAD)
1458 break;
1459 } while_each_thread(current, tsk);
1460 read_unlock(&tasklist_lock);
1461
1462notask:
1463 retval = wo->notask_error;
1464 if (!retval && !(wo->wo_flags & WNOHANG)) {
1465 retval = -ERESTARTSYS;
1466 if (!signal_pending(current)) {
1467 schedule();
1468 goto repeat;
1469 }
1470 }
1471end:
1472 __set_current_state(TASK_RUNNING);
1473 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1474 return retval;
1475}
1476
1477static struct pid *pidfd_get_pid(unsigned int fd)
1478{
1479 struct fd f;
1480 struct pid *pid;
1481
1482 f = fdget(fd);
1483 if (!f.file)
1484 return ERR_PTR(-EBADF);
1485
1486 pid = pidfd_pid(f.file);
1487 if (!IS_ERR(pid))
1488 get_pid(pid);
1489
1490 fdput(f);
1491 return pid;
1492}
1493
1494static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1495 int options, struct rusage *ru)
1496{
1497 struct wait_opts wo;
1498 struct pid *pid = NULL;
1499 enum pid_type type;
1500 long ret;
1501
1502 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1503 __WNOTHREAD|__WCLONE|__WALL))
1504 return -EINVAL;
1505 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1506 return -EINVAL;
1507
1508 switch (which) {
1509 case P_ALL:
1510 type = PIDTYPE_MAX;
1511 break;
1512 case P_PID:
1513 type = PIDTYPE_PID;
1514 if (upid <= 0)
1515 return -EINVAL;
1516
1517 pid = find_get_pid(upid);
1518 break;
1519 case P_PGID:
1520 type = PIDTYPE_PGID;
1521 if (upid < 0)
1522 return -EINVAL;
1523
1524 if (upid)
1525 pid = find_get_pid(upid);
1526 else
1527 pid = get_task_pid(current, PIDTYPE_PGID);
1528 break;
1529 case P_PIDFD:
1530 type = PIDTYPE_PID;
1531 if (upid < 0)
1532 return -EINVAL;
1533
1534 pid = pidfd_get_pid(upid);
1535 if (IS_ERR(pid))
1536 return PTR_ERR(pid);
1537 break;
1538 default:
1539 return -EINVAL;
1540 }
1541
1542 wo.wo_type = type;
1543 wo.wo_pid = pid;
1544 wo.wo_flags = options;
1545 wo.wo_info = infop;
1546 wo.wo_rusage = ru;
1547 ret = do_wait(&wo);
1548
1549 put_pid(pid);
1550 return ret;
1551}
1552
1553SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1554 infop, int, options, struct rusage __user *, ru)
1555{
1556 struct rusage r;
1557 struct waitid_info info = {.status = 0};
1558 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1559 int signo = 0;
1560
1561 if (err > 0) {
1562 signo = SIGCHLD;
1563 err = 0;
1564 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1565 return -EFAULT;
1566 }
1567 if (!infop)
1568 return err;
1569
1570 if (!user_write_access_begin(infop, sizeof(*infop)))
1571 return -EFAULT;
1572
1573 unsafe_put_user(signo, &infop->si_signo, Efault);
1574 unsafe_put_user(0, &infop->si_errno, Efault);
1575 unsafe_put_user(info.cause, &infop->si_code, Efault);
1576 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1577 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1578 unsafe_put_user(info.status, &infop->si_status, Efault);
1579 user_write_access_end();
1580 return err;
1581Efault:
1582 user_write_access_end();
1583 return -EFAULT;
1584}
1585
1586long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1587 struct rusage *ru)
1588{
1589 struct wait_opts wo;
1590 struct pid *pid = NULL;
1591 enum pid_type type;
1592 long ret;
1593
1594 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1595 __WNOTHREAD|__WCLONE|__WALL))
1596 return -EINVAL;
1597
1598 /* -INT_MIN is not defined */
1599 if (upid == INT_MIN)
1600 return -ESRCH;
1601
1602 if (upid == -1)
1603 type = PIDTYPE_MAX;
1604 else if (upid < 0) {
1605 type = PIDTYPE_PGID;
1606 pid = find_get_pid(-upid);
1607 } else if (upid == 0) {
1608 type = PIDTYPE_PGID;
1609 pid = get_task_pid(current, PIDTYPE_PGID);
1610 } else /* upid > 0 */ {
1611 type = PIDTYPE_PID;
1612 pid = find_get_pid(upid);
1613 }
1614
1615 wo.wo_type = type;
1616 wo.wo_pid = pid;
1617 wo.wo_flags = options | WEXITED;
1618 wo.wo_info = NULL;
1619 wo.wo_stat = 0;
1620 wo.wo_rusage = ru;
1621 ret = do_wait(&wo);
1622 put_pid(pid);
1623 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1624 ret = -EFAULT;
1625
1626 return ret;
1627}
1628
1629int kernel_wait(pid_t pid, int *stat)
1630{
1631 struct wait_opts wo = {
1632 .wo_type = PIDTYPE_PID,
1633 .wo_pid = find_get_pid(pid),
1634 .wo_flags = WEXITED,
1635 };
1636 int ret;
1637
1638 ret = do_wait(&wo);
1639 if (ret > 0 && wo.wo_stat)
1640 *stat = wo.wo_stat;
1641 put_pid(wo.wo_pid);
1642 return ret;
1643}
1644
1645SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1646 int, options, struct rusage __user *, ru)
1647{
1648 struct rusage r;
1649 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1650
1651 if (err > 0) {
1652 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1653 return -EFAULT;
1654 }
1655 return err;
1656}
1657
1658#ifdef __ARCH_WANT_SYS_WAITPID
1659
1660/*
1661 * sys_waitpid() remains for compatibility. waitpid() should be
1662 * implemented by calling sys_wait4() from libc.a.
1663 */
1664SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1665{
1666 return kernel_wait4(pid, stat_addr, options, NULL);
1667}
1668
1669#endif
1670
1671#ifdef CONFIG_COMPAT
1672COMPAT_SYSCALL_DEFINE4(wait4,
1673 compat_pid_t, pid,
1674 compat_uint_t __user *, stat_addr,
1675 int, options,
1676 struct compat_rusage __user *, ru)
1677{
1678 struct rusage r;
1679 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1680 if (err > 0) {
1681 if (ru && put_compat_rusage(&r, ru))
1682 return -EFAULT;
1683 }
1684 return err;
1685}
1686
1687COMPAT_SYSCALL_DEFINE5(waitid,
1688 int, which, compat_pid_t, pid,
1689 struct compat_siginfo __user *, infop, int, options,
1690 struct compat_rusage __user *, uru)
1691{
1692 struct rusage ru;
1693 struct waitid_info info = {.status = 0};
1694 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1695 int signo = 0;
1696 if (err > 0) {
1697 signo = SIGCHLD;
1698 err = 0;
1699 if (uru) {
1700 /* kernel_waitid() overwrites everything in ru */
1701 if (COMPAT_USE_64BIT_TIME)
1702 err = copy_to_user(uru, &ru, sizeof(ru));
1703 else
1704 err = put_compat_rusage(&ru, uru);
1705 if (err)
1706 return -EFAULT;
1707 }
1708 }
1709
1710 if (!infop)
1711 return err;
1712
1713 if (!user_write_access_begin(infop, sizeof(*infop)))
1714 return -EFAULT;
1715
1716 unsafe_put_user(signo, &infop->si_signo, Efault);
1717 unsafe_put_user(0, &infop->si_errno, Efault);
1718 unsafe_put_user(info.cause, &infop->si_code, Efault);
1719 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1720 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1721 unsafe_put_user(info.status, &infop->si_status, Efault);
1722 user_write_access_end();
1723 return err;
1724Efault:
1725 user_write_access_end();
1726 return -EFAULT;
1727}
1728#endif
1729
1730/**
1731 * thread_group_exited - check that a thread group has exited
1732 * @pid: tgid of thread group to be checked.
1733 *
1734 * Test if the thread group represented by tgid has exited (all
1735 * threads are zombies, dead or completely gone).
1736 *
1737 * Return: true if the thread group has exited. false otherwise.
1738 */
1739bool thread_group_exited(struct pid *pid)
1740{
1741 struct task_struct *task;
1742 bool exited;
1743
1744 rcu_read_lock();
1745 task = pid_task(pid, PIDTYPE_PID);
1746 exited = !task ||
1747 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1748 rcu_read_unlock();
1749
1750 return exited;
1751}
1752EXPORT_SYMBOL(thread_group_exited);
1753
1754__weak void abort(void)
1755{
1756 BUG();
1757
1758 /* if that doesn't kill us, halt */
1759 panic("Oops failed to kill thread");
1760}
1761EXPORT_SYMBOL(abort);