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