Loading...
1/*
2 * linux/kernel/exit.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/mm.h>
8#include <linux/slab.h>
9#include <linux/interrupt.h>
10#include <linux/module.h>
11#include <linux/capability.h>
12#include <linux/completion.h>
13#include <linux/personality.h>
14#include <linux/tty.h>
15#include <linux/iocontext.h>
16#include <linux/key.h>
17#include <linux/security.h>
18#include <linux/cpu.h>
19#include <linux/acct.h>
20#include <linux/tsacct_kern.h>
21#include <linux/file.h>
22#include <linux/fdtable.h>
23#include <linux/freezer.h>
24#include <linux/binfmts.h>
25#include <linux/nsproxy.h>
26#include <linux/pid_namespace.h>
27#include <linux/ptrace.h>
28#include <linux/profile.h>
29#include <linux/mount.h>
30#include <linux/proc_fs.h>
31#include <linux/kthread.h>
32#include <linux/mempolicy.h>
33#include <linux/taskstats_kern.h>
34#include <linux/delayacct.h>
35#include <linux/cgroup.h>
36#include <linux/syscalls.h>
37#include <linux/signal.h>
38#include <linux/posix-timers.h>
39#include <linux/cn_proc.h>
40#include <linux/mutex.h>
41#include <linux/futex.h>
42#include <linux/pipe_fs_i.h>
43#include <linux/audit.h> /* for audit_free() */
44#include <linux/resource.h>
45#include <linux/blkdev.h>
46#include <linux/task_io_accounting_ops.h>
47#include <linux/tracehook.h>
48#include <linux/fs_struct.h>
49#include <linux/init_task.h>
50#include <linux/perf_event.h>
51#include <trace/events/sched.h>
52#include <linux/hw_breakpoint.h>
53#include <linux/oom.h>
54#include <linux/writeback.h>
55#include <linux/shm.h>
56#include <linux/kcov.h>
57#include <linux/random.h>
58
59#include <linux/uaccess.h>
60#include <asm/unistd.h>
61#include <asm/pgtable.h>
62#include <asm/mmu_context.h>
63
64static void __unhash_process(struct task_struct *p, bool group_dead)
65{
66 nr_threads--;
67 detach_pid(p, PIDTYPE_PID);
68 if (group_dead) {
69 detach_pid(p, PIDTYPE_PGID);
70 detach_pid(p, PIDTYPE_SID);
71
72 list_del_rcu(&p->tasks);
73 list_del_init(&p->sibling);
74 __this_cpu_dec(process_counts);
75 }
76 list_del_rcu(&p->thread_group);
77 list_del_rcu(&p->thread_node);
78}
79
80/*
81 * This function expects the tasklist_lock write-locked.
82 */
83static void __exit_signal(struct task_struct *tsk)
84{
85 struct signal_struct *sig = tsk->signal;
86 bool group_dead = thread_group_leader(tsk);
87 struct sighand_struct *sighand;
88 struct tty_struct *uninitialized_var(tty);
89 cputime_t utime, stime;
90
91 sighand = rcu_dereference_check(tsk->sighand,
92 lockdep_tasklist_lock_is_held());
93 spin_lock(&sighand->siglock);
94
95#ifdef CONFIG_POSIX_TIMERS
96 posix_cpu_timers_exit(tsk);
97 if (group_dead) {
98 posix_cpu_timers_exit_group(tsk);
99 } else {
100 /*
101 * This can only happen if the caller is de_thread().
102 * FIXME: this is the temporary hack, we should teach
103 * posix-cpu-timers to handle this case correctly.
104 */
105 if (unlikely(has_group_leader_pid(tsk)))
106 posix_cpu_timers_exit_group(tsk);
107 }
108#endif
109
110 if (group_dead) {
111 tty = sig->tty;
112 sig->tty = NULL;
113 } else {
114 /*
115 * If there is any task waiting for the group exit
116 * then notify it:
117 */
118 if (sig->notify_count > 0 && !--sig->notify_count)
119 wake_up_process(sig->group_exit_task);
120
121 if (tsk == sig->curr_target)
122 sig->curr_target = next_thread(tsk);
123 }
124
125 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
126 sizeof(unsigned long long));
127
128 /*
129 * Accumulate here the counters for all threads as they die. We could
130 * skip the group leader because it is the last user of signal_struct,
131 * but we want to avoid the race with thread_group_cputime() which can
132 * see the empty ->thread_head list.
133 */
134 task_cputime(tsk, &utime, &stime);
135 write_seqlock(&sig->stats_lock);
136 sig->utime += utime;
137 sig->stime += stime;
138 sig->gtime += task_gtime(tsk);
139 sig->min_flt += tsk->min_flt;
140 sig->maj_flt += tsk->maj_flt;
141 sig->nvcsw += tsk->nvcsw;
142 sig->nivcsw += tsk->nivcsw;
143 sig->inblock += task_io_get_inblock(tsk);
144 sig->oublock += task_io_get_oublock(tsk);
145 task_io_accounting_add(&sig->ioac, &tsk->ioac);
146 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
147 sig->nr_threads--;
148 __unhash_process(tsk, group_dead);
149 write_sequnlock(&sig->stats_lock);
150
151 /*
152 * Do this under ->siglock, we can race with another thread
153 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
154 */
155 flush_sigqueue(&tsk->pending);
156 tsk->sighand = NULL;
157 spin_unlock(&sighand->siglock);
158
159 __cleanup_sighand(sighand);
160 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
161 if (group_dead) {
162 flush_sigqueue(&sig->shared_pending);
163 tty_kref_put(tty);
164 }
165}
166
167static void delayed_put_task_struct(struct rcu_head *rhp)
168{
169 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
170
171 perf_event_delayed_put(tsk);
172 trace_sched_process_free(tsk);
173 put_task_struct(tsk);
174}
175
176
177void release_task(struct task_struct *p)
178{
179 struct task_struct *leader;
180 int zap_leader;
181repeat:
182 /* don't need to get the RCU readlock here - the process is dead and
183 * can't be modifying its own credentials. But shut RCU-lockdep up */
184 rcu_read_lock();
185 atomic_dec(&__task_cred(p)->user->processes);
186 rcu_read_unlock();
187
188 proc_flush_task(p);
189
190 write_lock_irq(&tasklist_lock);
191 ptrace_release_task(p);
192 __exit_signal(p);
193
194 /*
195 * If we are the last non-leader member of the thread
196 * group, and the leader is zombie, then notify the
197 * group leader's parent process. (if it wants notification.)
198 */
199 zap_leader = 0;
200 leader = p->group_leader;
201 if (leader != p && thread_group_empty(leader)
202 && leader->exit_state == EXIT_ZOMBIE) {
203 /*
204 * If we were the last child thread and the leader has
205 * exited already, and the leader's parent ignores SIGCHLD,
206 * then we are the one who should release the leader.
207 */
208 zap_leader = do_notify_parent(leader, leader->exit_signal);
209 if (zap_leader)
210 leader->exit_state = EXIT_DEAD;
211 }
212
213 write_unlock_irq(&tasklist_lock);
214 release_thread(p);
215 call_rcu(&p->rcu, delayed_put_task_struct);
216
217 p = leader;
218 if (unlikely(zap_leader))
219 goto repeat;
220}
221
222/*
223 * Note that if this function returns a valid task_struct pointer (!NULL)
224 * task->usage must remain >0 for the duration of the RCU critical section.
225 */
226struct task_struct *task_rcu_dereference(struct task_struct **ptask)
227{
228 struct sighand_struct *sighand;
229 struct task_struct *task;
230
231 /*
232 * We need to verify that release_task() was not called and thus
233 * delayed_put_task_struct() can't run and drop the last reference
234 * before rcu_read_unlock(). We check task->sighand != NULL,
235 * but we can read the already freed and reused memory.
236 */
237retry:
238 task = rcu_dereference(*ptask);
239 if (!task)
240 return NULL;
241
242 probe_kernel_address(&task->sighand, sighand);
243
244 /*
245 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
246 * was already freed we can not miss the preceding update of this
247 * pointer.
248 */
249 smp_rmb();
250 if (unlikely(task != READ_ONCE(*ptask)))
251 goto retry;
252
253 /*
254 * We've re-checked that "task == *ptask", now we have two different
255 * cases:
256 *
257 * 1. This is actually the same task/task_struct. In this case
258 * sighand != NULL tells us it is still alive.
259 *
260 * 2. This is another task which got the same memory for task_struct.
261 * We can't know this of course, and we can not trust
262 * sighand != NULL.
263 *
264 * In this case we actually return a random value, but this is
265 * correct.
266 *
267 * If we return NULL - we can pretend that we actually noticed that
268 * *ptask was updated when the previous task has exited. Or pretend
269 * that probe_slab_address(&sighand) reads NULL.
270 *
271 * If we return the new task (because sighand is not NULL for any
272 * reason) - this is fine too. This (new) task can't go away before
273 * another gp pass.
274 *
275 * And note: We could even eliminate the false positive if re-read
276 * task->sighand once again to avoid the falsely NULL. But this case
277 * is very unlikely so we don't care.
278 */
279 if (!sighand)
280 return NULL;
281
282 return task;
283}
284
285struct task_struct *try_get_task_struct(struct task_struct **ptask)
286{
287 struct task_struct *task;
288
289 rcu_read_lock();
290 task = task_rcu_dereference(ptask);
291 if (task)
292 get_task_struct(task);
293 rcu_read_unlock();
294
295 return task;
296}
297
298/*
299 * Determine if a process group is "orphaned", according to the POSIX
300 * definition in 2.2.2.52. Orphaned process groups are not to be affected
301 * by terminal-generated stop signals. Newly orphaned process groups are
302 * to receive a SIGHUP and a SIGCONT.
303 *
304 * "I ask you, have you ever known what it is to be an orphan?"
305 */
306static int will_become_orphaned_pgrp(struct pid *pgrp,
307 struct task_struct *ignored_task)
308{
309 struct task_struct *p;
310
311 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
312 if ((p == ignored_task) ||
313 (p->exit_state && thread_group_empty(p)) ||
314 is_global_init(p->real_parent))
315 continue;
316
317 if (task_pgrp(p->real_parent) != pgrp &&
318 task_session(p->real_parent) == task_session(p))
319 return 0;
320 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
321
322 return 1;
323}
324
325int is_current_pgrp_orphaned(void)
326{
327 int retval;
328
329 read_lock(&tasklist_lock);
330 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
331 read_unlock(&tasklist_lock);
332
333 return retval;
334}
335
336static bool has_stopped_jobs(struct pid *pgrp)
337{
338 struct task_struct *p;
339
340 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
341 if (p->signal->flags & SIGNAL_STOP_STOPPED)
342 return true;
343 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
344
345 return false;
346}
347
348/*
349 * Check to see if any process groups have become orphaned as
350 * a result of our exiting, and if they have any stopped jobs,
351 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
352 */
353static void
354kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
355{
356 struct pid *pgrp = task_pgrp(tsk);
357 struct task_struct *ignored_task = tsk;
358
359 if (!parent)
360 /* exit: our father is in a different pgrp than
361 * we are and we were the only connection outside.
362 */
363 parent = tsk->real_parent;
364 else
365 /* reparent: our child is in a different pgrp than
366 * we are, and it was the only connection outside.
367 */
368 ignored_task = NULL;
369
370 if (task_pgrp(parent) != pgrp &&
371 task_session(parent) == task_session(tsk) &&
372 will_become_orphaned_pgrp(pgrp, ignored_task) &&
373 has_stopped_jobs(pgrp)) {
374 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
375 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
376 }
377}
378
379#ifdef CONFIG_MEMCG
380/*
381 * A task is exiting. If it owned this mm, find a new owner for the mm.
382 */
383void mm_update_next_owner(struct mm_struct *mm)
384{
385 struct task_struct *c, *g, *p = current;
386
387retry:
388 /*
389 * If the exiting or execing task is not the owner, it's
390 * someone else's problem.
391 */
392 if (mm->owner != p)
393 return;
394 /*
395 * The current owner is exiting/execing and there are no other
396 * candidates. Do not leave the mm pointing to a possibly
397 * freed task structure.
398 */
399 if (atomic_read(&mm->mm_users) <= 1) {
400 mm->owner = NULL;
401 return;
402 }
403
404 read_lock(&tasklist_lock);
405 /*
406 * Search in the children
407 */
408 list_for_each_entry(c, &p->children, sibling) {
409 if (c->mm == mm)
410 goto assign_new_owner;
411 }
412
413 /*
414 * Search in the siblings
415 */
416 list_for_each_entry(c, &p->real_parent->children, sibling) {
417 if (c->mm == mm)
418 goto assign_new_owner;
419 }
420
421 /*
422 * Search through everything else, we should not get here often.
423 */
424 for_each_process(g) {
425 if (g->flags & PF_KTHREAD)
426 continue;
427 for_each_thread(g, c) {
428 if (c->mm == mm)
429 goto assign_new_owner;
430 if (c->mm)
431 break;
432 }
433 }
434 read_unlock(&tasklist_lock);
435 /*
436 * We found no owner yet mm_users > 1: this implies that we are
437 * most likely racing with swapoff (try_to_unuse()) or /proc or
438 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
439 */
440 mm->owner = NULL;
441 return;
442
443assign_new_owner:
444 BUG_ON(c == p);
445 get_task_struct(c);
446 /*
447 * The task_lock protects c->mm from changing.
448 * We always want mm->owner->mm == mm
449 */
450 task_lock(c);
451 /*
452 * Delay read_unlock() till we have the task_lock()
453 * to ensure that c does not slip away underneath us
454 */
455 read_unlock(&tasklist_lock);
456 if (c->mm != mm) {
457 task_unlock(c);
458 put_task_struct(c);
459 goto retry;
460 }
461 mm->owner = c;
462 task_unlock(c);
463 put_task_struct(c);
464}
465#endif /* CONFIG_MEMCG */
466
467/*
468 * Turn us into a lazy TLB process if we
469 * aren't already..
470 */
471static void exit_mm(struct task_struct *tsk)
472{
473 struct mm_struct *mm = tsk->mm;
474 struct core_state *core_state;
475
476 mm_release(tsk, mm);
477 if (!mm)
478 return;
479 sync_mm_rss(mm);
480 /*
481 * Serialize with any possible pending coredump.
482 * We must hold mmap_sem around checking core_state
483 * and clearing tsk->mm. The core-inducing thread
484 * will increment ->nr_threads for each thread in the
485 * group with ->mm != NULL.
486 */
487 down_read(&mm->mmap_sem);
488 core_state = mm->core_state;
489 if (core_state) {
490 struct core_thread self;
491
492 up_read(&mm->mmap_sem);
493
494 self.task = tsk;
495 self.next = xchg(&core_state->dumper.next, &self);
496 /*
497 * Implies mb(), the result of xchg() must be visible
498 * to core_state->dumper.
499 */
500 if (atomic_dec_and_test(&core_state->nr_threads))
501 complete(&core_state->startup);
502
503 for (;;) {
504 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
505 if (!self.task) /* see coredump_finish() */
506 break;
507 freezable_schedule();
508 }
509 __set_task_state(tsk, TASK_RUNNING);
510 down_read(&mm->mmap_sem);
511 }
512 atomic_inc(&mm->mm_count);
513 BUG_ON(mm != tsk->active_mm);
514 /* more a memory barrier than a real lock */
515 task_lock(tsk);
516 tsk->mm = NULL;
517 up_read(&mm->mmap_sem);
518 enter_lazy_tlb(mm, current);
519 task_unlock(tsk);
520 mm_update_next_owner(mm);
521 mmput(mm);
522 if (test_thread_flag(TIF_MEMDIE))
523 exit_oom_victim();
524}
525
526static struct task_struct *find_alive_thread(struct task_struct *p)
527{
528 struct task_struct *t;
529
530 for_each_thread(p, t) {
531 if (!(t->flags & PF_EXITING))
532 return t;
533 }
534 return NULL;
535}
536
537static struct task_struct *find_child_reaper(struct task_struct *father)
538 __releases(&tasklist_lock)
539 __acquires(&tasklist_lock)
540{
541 struct pid_namespace *pid_ns = task_active_pid_ns(father);
542 struct task_struct *reaper = pid_ns->child_reaper;
543
544 if (likely(reaper != father))
545 return reaper;
546
547 reaper = find_alive_thread(father);
548 if (reaper) {
549 pid_ns->child_reaper = reaper;
550 return reaper;
551 }
552
553 write_unlock_irq(&tasklist_lock);
554 if (unlikely(pid_ns == &init_pid_ns)) {
555 panic("Attempted to kill init! exitcode=0x%08x\n",
556 father->signal->group_exit_code ?: father->exit_code);
557 }
558 zap_pid_ns_processes(pid_ns);
559 write_lock_irq(&tasklist_lock);
560
561 return father;
562}
563
564/*
565 * When we die, we re-parent all our children, and try to:
566 * 1. give them to another thread in our thread group, if such a member exists
567 * 2. give it to the first ancestor process which prctl'd itself as a
568 * child_subreaper for its children (like a service manager)
569 * 3. give it to the init process (PID 1) in our pid namespace
570 */
571static struct task_struct *find_new_reaper(struct task_struct *father,
572 struct task_struct *child_reaper)
573{
574 struct task_struct *thread, *reaper;
575
576 thread = find_alive_thread(father);
577 if (thread)
578 return thread;
579
580 if (father->signal->has_child_subreaper) {
581 /*
582 * Find the first ->is_child_subreaper ancestor in our pid_ns.
583 * We start from father to ensure we can not look into another
584 * namespace, this is safe because all its threads are dead.
585 */
586 for (reaper = father;
587 !same_thread_group(reaper, child_reaper);
588 reaper = reaper->real_parent) {
589 /* call_usermodehelper() descendants need this check */
590 if (reaper == &init_task)
591 break;
592 if (!reaper->signal->is_child_subreaper)
593 continue;
594 thread = find_alive_thread(reaper);
595 if (thread)
596 return thread;
597 }
598 }
599
600 return child_reaper;
601}
602
603/*
604* Any that need to be release_task'd are put on the @dead list.
605 */
606static void reparent_leader(struct task_struct *father, struct task_struct *p,
607 struct list_head *dead)
608{
609 if (unlikely(p->exit_state == EXIT_DEAD))
610 return;
611
612 /* We don't want people slaying init. */
613 p->exit_signal = SIGCHLD;
614
615 /* If it has exited notify the new parent about this child's death. */
616 if (!p->ptrace &&
617 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
618 if (do_notify_parent(p, p->exit_signal)) {
619 p->exit_state = EXIT_DEAD;
620 list_add(&p->ptrace_entry, dead);
621 }
622 }
623
624 kill_orphaned_pgrp(p, father);
625}
626
627/*
628 * This does two things:
629 *
630 * A. Make init inherit all the child processes
631 * B. Check to see if any process groups have become orphaned
632 * as a result of our exiting, and if they have any stopped
633 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
634 */
635static void forget_original_parent(struct task_struct *father,
636 struct list_head *dead)
637{
638 struct task_struct *p, *t, *reaper;
639
640 if (unlikely(!list_empty(&father->ptraced)))
641 exit_ptrace(father, dead);
642
643 /* Can drop and reacquire tasklist_lock */
644 reaper = find_child_reaper(father);
645 if (list_empty(&father->children))
646 return;
647
648 reaper = find_new_reaper(father, reaper);
649 list_for_each_entry(p, &father->children, sibling) {
650 for_each_thread(p, t) {
651 t->real_parent = reaper;
652 BUG_ON((!t->ptrace) != (t->parent == father));
653 if (likely(!t->ptrace))
654 t->parent = t->real_parent;
655 if (t->pdeath_signal)
656 group_send_sig_info(t->pdeath_signal,
657 SEND_SIG_NOINFO, t);
658 }
659 /*
660 * If this is a threaded reparent there is no need to
661 * notify anyone anything has happened.
662 */
663 if (!same_thread_group(reaper, father))
664 reparent_leader(father, p, dead);
665 }
666 list_splice_tail_init(&father->children, &reaper->children);
667}
668
669/*
670 * Send signals to all our closest relatives so that they know
671 * to properly mourn us..
672 */
673static void exit_notify(struct task_struct *tsk, int group_dead)
674{
675 bool autoreap;
676 struct task_struct *p, *n;
677 LIST_HEAD(dead);
678
679 write_lock_irq(&tasklist_lock);
680 forget_original_parent(tsk, &dead);
681
682 if (group_dead)
683 kill_orphaned_pgrp(tsk->group_leader, NULL);
684
685 if (unlikely(tsk->ptrace)) {
686 int sig = thread_group_leader(tsk) &&
687 thread_group_empty(tsk) &&
688 !ptrace_reparented(tsk) ?
689 tsk->exit_signal : SIGCHLD;
690 autoreap = do_notify_parent(tsk, sig);
691 } else if (thread_group_leader(tsk)) {
692 autoreap = thread_group_empty(tsk) &&
693 do_notify_parent(tsk, tsk->exit_signal);
694 } else {
695 autoreap = true;
696 }
697
698 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
699 if (tsk->exit_state == EXIT_DEAD)
700 list_add(&tsk->ptrace_entry, &dead);
701
702 /* mt-exec, de_thread() is waiting for group leader */
703 if (unlikely(tsk->signal->notify_count < 0))
704 wake_up_process(tsk->signal->group_exit_task);
705 write_unlock_irq(&tasklist_lock);
706
707 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
708 list_del_init(&p->ptrace_entry);
709 release_task(p);
710 }
711}
712
713#ifdef CONFIG_DEBUG_STACK_USAGE
714static void check_stack_usage(void)
715{
716 static DEFINE_SPINLOCK(low_water_lock);
717 static int lowest_to_date = THREAD_SIZE;
718 unsigned long free;
719
720 free = stack_not_used(current);
721
722 if (free >= lowest_to_date)
723 return;
724
725 spin_lock(&low_water_lock);
726 if (free < lowest_to_date) {
727 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
728 current->comm, task_pid_nr(current), free);
729 lowest_to_date = free;
730 }
731 spin_unlock(&low_water_lock);
732}
733#else
734static inline void check_stack_usage(void) {}
735#endif
736
737void __noreturn do_exit(long code)
738{
739 struct task_struct *tsk = current;
740 int group_dead;
741 TASKS_RCU(int tasks_rcu_i);
742
743 profile_task_exit(tsk);
744 kcov_task_exit(tsk);
745
746 WARN_ON(blk_needs_flush_plug(tsk));
747
748 if (unlikely(in_interrupt()))
749 panic("Aiee, killing interrupt handler!");
750 if (unlikely(!tsk->pid))
751 panic("Attempted to kill the idle task!");
752
753 /*
754 * If do_exit is called because this processes oopsed, it's possible
755 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
756 * continuing. Amongst other possible reasons, this is to prevent
757 * mm_release()->clear_child_tid() from writing to a user-controlled
758 * kernel address.
759 */
760 set_fs(USER_DS);
761
762 ptrace_event(PTRACE_EVENT_EXIT, code);
763
764 validate_creds_for_do_exit(tsk);
765
766 /*
767 * We're taking recursive faults here in do_exit. Safest is to just
768 * leave this task alone and wait for reboot.
769 */
770 if (unlikely(tsk->flags & PF_EXITING)) {
771 pr_alert("Fixing recursive fault but reboot is needed!\n");
772 /*
773 * We can do this unlocked here. The futex code uses
774 * this flag just to verify whether the pi state
775 * cleanup has been done or not. In the worst case it
776 * loops once more. We pretend that the cleanup was
777 * done as there is no way to return. Either the
778 * OWNER_DIED bit is set by now or we push the blocked
779 * task into the wait for ever nirwana as well.
780 */
781 tsk->flags |= PF_EXITPIDONE;
782 set_current_state(TASK_UNINTERRUPTIBLE);
783 schedule();
784 }
785
786 exit_signals(tsk); /* sets PF_EXITING */
787 /*
788 * Ensure that all new tsk->pi_lock acquisitions must observe
789 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
790 */
791 smp_mb();
792 /*
793 * Ensure that we must observe the pi_state in exit_mm() ->
794 * mm_release() -> exit_pi_state_list().
795 */
796 raw_spin_unlock_wait(&tsk->pi_lock);
797
798 if (unlikely(in_atomic())) {
799 pr_info("note: %s[%d] exited with preempt_count %d\n",
800 current->comm, task_pid_nr(current),
801 preempt_count());
802 preempt_count_set(PREEMPT_ENABLED);
803 }
804
805 /* sync mm's RSS info before statistics gathering */
806 if (tsk->mm)
807 sync_mm_rss(tsk->mm);
808 acct_update_integrals(tsk);
809 group_dead = atomic_dec_and_test(&tsk->signal->live);
810 if (group_dead) {
811#ifdef CONFIG_POSIX_TIMERS
812 hrtimer_cancel(&tsk->signal->real_timer);
813 exit_itimers(tsk->signal);
814#endif
815 if (tsk->mm)
816 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
817 }
818 acct_collect(code, group_dead);
819 if (group_dead)
820 tty_audit_exit();
821 audit_free(tsk);
822
823 tsk->exit_code = code;
824 taskstats_exit(tsk, group_dead);
825
826 exit_mm(tsk);
827
828 if (group_dead)
829 acct_process();
830 trace_sched_process_exit(tsk);
831
832 exit_sem(tsk);
833 exit_shm(tsk);
834 exit_files(tsk);
835 exit_fs(tsk);
836 if (group_dead)
837 disassociate_ctty(1);
838 exit_task_namespaces(tsk);
839 exit_task_work(tsk);
840 exit_thread(tsk);
841
842 /*
843 * Flush inherited counters to the parent - before the parent
844 * gets woken up by child-exit notifications.
845 *
846 * because of cgroup mode, must be called before cgroup_exit()
847 */
848 perf_event_exit_task(tsk);
849
850 sched_autogroup_exit_task(tsk);
851 cgroup_exit(tsk);
852
853 /*
854 * FIXME: do that only when needed, using sched_exit tracepoint
855 */
856 flush_ptrace_hw_breakpoint(tsk);
857
858 TASKS_RCU(preempt_disable());
859 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
860 TASKS_RCU(preempt_enable());
861 exit_notify(tsk, group_dead);
862 proc_exit_connector(tsk);
863 mpol_put_task_policy(tsk);
864#ifdef CONFIG_FUTEX
865 if (unlikely(current->pi_state_cache))
866 kfree(current->pi_state_cache);
867#endif
868 /*
869 * Make sure we are holding no locks:
870 */
871 debug_check_no_locks_held();
872 /*
873 * We can do this unlocked here. The futex code uses this flag
874 * just to verify whether the pi state cleanup has been done
875 * or not. In the worst case it loops once more.
876 */
877 tsk->flags |= PF_EXITPIDONE;
878
879 if (tsk->io_context)
880 exit_io_context(tsk);
881
882 if (tsk->splice_pipe)
883 free_pipe_info(tsk->splice_pipe);
884
885 if (tsk->task_frag.page)
886 put_page(tsk->task_frag.page);
887
888 validate_creds_for_do_exit(tsk);
889
890 check_stack_usage();
891 preempt_disable();
892 if (tsk->nr_dirtied)
893 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
894 exit_rcu();
895 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
896
897 do_task_dead();
898}
899EXPORT_SYMBOL_GPL(do_exit);
900
901void complete_and_exit(struct completion *comp, long code)
902{
903 if (comp)
904 complete(comp);
905
906 do_exit(code);
907}
908EXPORT_SYMBOL(complete_and_exit);
909
910SYSCALL_DEFINE1(exit, int, error_code)
911{
912 do_exit((error_code&0xff)<<8);
913}
914
915/*
916 * Take down every thread in the group. This is called by fatal signals
917 * as well as by sys_exit_group (below).
918 */
919void
920do_group_exit(int exit_code)
921{
922 struct signal_struct *sig = current->signal;
923
924 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
925
926 if (signal_group_exit(sig))
927 exit_code = sig->group_exit_code;
928 else if (!thread_group_empty(current)) {
929 struct sighand_struct *const sighand = current->sighand;
930
931 spin_lock_irq(&sighand->siglock);
932 if (signal_group_exit(sig))
933 /* Another thread got here before we took the lock. */
934 exit_code = sig->group_exit_code;
935 else {
936 sig->group_exit_code = exit_code;
937 sig->flags = SIGNAL_GROUP_EXIT;
938 zap_other_threads(current);
939 }
940 spin_unlock_irq(&sighand->siglock);
941 }
942
943 do_exit(exit_code);
944 /* NOTREACHED */
945}
946
947/*
948 * this kills every thread in the thread group. Note that any externally
949 * wait4()-ing process will get the correct exit code - even if this
950 * thread is not the thread group leader.
951 */
952SYSCALL_DEFINE1(exit_group, int, error_code)
953{
954 do_group_exit((error_code & 0xff) << 8);
955 /* NOTREACHED */
956 return 0;
957}
958
959struct wait_opts {
960 enum pid_type wo_type;
961 int wo_flags;
962 struct pid *wo_pid;
963
964 struct siginfo __user *wo_info;
965 int __user *wo_stat;
966 struct rusage __user *wo_rusage;
967
968 wait_queue_t child_wait;
969 int notask_error;
970};
971
972static inline
973struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
974{
975 if (type != PIDTYPE_PID)
976 task = task->group_leader;
977 return task->pids[type].pid;
978}
979
980static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
981{
982 return wo->wo_type == PIDTYPE_MAX ||
983 task_pid_type(p, wo->wo_type) == wo->wo_pid;
984}
985
986static int
987eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
988{
989 if (!eligible_pid(wo, p))
990 return 0;
991
992 /*
993 * Wait for all children (clone and not) if __WALL is set or
994 * if it is traced by us.
995 */
996 if (ptrace || (wo->wo_flags & __WALL))
997 return 1;
998
999 /*
1000 * Otherwise, wait for clone children *only* if __WCLONE is set;
1001 * otherwise, wait for non-clone children *only*.
1002 *
1003 * Note: a "clone" child here is one that reports to its parent
1004 * using a signal other than SIGCHLD, or a non-leader thread which
1005 * we can only see if it is traced by us.
1006 */
1007 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1008 return 0;
1009
1010 return 1;
1011}
1012
1013static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1014 pid_t pid, uid_t uid, int why, int status)
1015{
1016 struct siginfo __user *infop;
1017 int retval = wo->wo_rusage
1018 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1019
1020 put_task_struct(p);
1021 infop = wo->wo_info;
1022 if (infop) {
1023 if (!retval)
1024 retval = put_user(SIGCHLD, &infop->si_signo);
1025 if (!retval)
1026 retval = put_user(0, &infop->si_errno);
1027 if (!retval)
1028 retval = put_user((short)why, &infop->si_code);
1029 if (!retval)
1030 retval = put_user(pid, &infop->si_pid);
1031 if (!retval)
1032 retval = put_user(uid, &infop->si_uid);
1033 if (!retval)
1034 retval = put_user(status, &infop->si_status);
1035 }
1036 if (!retval)
1037 retval = pid;
1038 return retval;
1039}
1040
1041/*
1042 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1043 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1044 * the lock and this task is uninteresting. If we return nonzero, we have
1045 * released the lock and the system call should return.
1046 */
1047static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1048{
1049 int state, retval, status;
1050 pid_t pid = task_pid_vnr(p);
1051 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1052 struct siginfo __user *infop;
1053
1054 if (!likely(wo->wo_flags & WEXITED))
1055 return 0;
1056
1057 if (unlikely(wo->wo_flags & WNOWAIT)) {
1058 int exit_code = p->exit_code;
1059 int why;
1060
1061 get_task_struct(p);
1062 read_unlock(&tasklist_lock);
1063 sched_annotate_sleep();
1064
1065 if ((exit_code & 0x7f) == 0) {
1066 why = CLD_EXITED;
1067 status = exit_code >> 8;
1068 } else {
1069 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1070 status = exit_code & 0x7f;
1071 }
1072 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1073 }
1074 /*
1075 * Move the task's state to DEAD/TRACE, only one thread can do this.
1076 */
1077 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1078 EXIT_TRACE : EXIT_DEAD;
1079 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1080 return 0;
1081 /*
1082 * We own this thread, nobody else can reap it.
1083 */
1084 read_unlock(&tasklist_lock);
1085 sched_annotate_sleep();
1086
1087 /*
1088 * Check thread_group_leader() to exclude the traced sub-threads.
1089 */
1090 if (state == EXIT_DEAD && thread_group_leader(p)) {
1091 struct signal_struct *sig = p->signal;
1092 struct signal_struct *psig = current->signal;
1093 unsigned long maxrss;
1094 cputime_t tgutime, tgstime;
1095
1096 /*
1097 * The resource counters for the group leader are in its
1098 * own task_struct. Those for dead threads in the group
1099 * are in its signal_struct, as are those for the child
1100 * processes it has previously reaped. All these
1101 * accumulate in the parent's signal_struct c* fields.
1102 *
1103 * We don't bother to take a lock here to protect these
1104 * p->signal fields because the whole thread group is dead
1105 * and nobody can change them.
1106 *
1107 * psig->stats_lock also protects us from our sub-theads
1108 * which can reap other children at the same time. Until
1109 * we change k_getrusage()-like users to rely on this lock
1110 * we have to take ->siglock as well.
1111 *
1112 * We use thread_group_cputime_adjusted() to get times for
1113 * the thread group, which consolidates times for all threads
1114 * in the group including the group leader.
1115 */
1116 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1117 spin_lock_irq(¤t->sighand->siglock);
1118 write_seqlock(&psig->stats_lock);
1119 psig->cutime += tgutime + sig->cutime;
1120 psig->cstime += tgstime + sig->cstime;
1121 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1122 psig->cmin_flt +=
1123 p->min_flt + sig->min_flt + sig->cmin_flt;
1124 psig->cmaj_flt +=
1125 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1126 psig->cnvcsw +=
1127 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1128 psig->cnivcsw +=
1129 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1130 psig->cinblock +=
1131 task_io_get_inblock(p) +
1132 sig->inblock + sig->cinblock;
1133 psig->coublock +=
1134 task_io_get_oublock(p) +
1135 sig->oublock + sig->coublock;
1136 maxrss = max(sig->maxrss, sig->cmaxrss);
1137 if (psig->cmaxrss < maxrss)
1138 psig->cmaxrss = maxrss;
1139 task_io_accounting_add(&psig->ioac, &p->ioac);
1140 task_io_accounting_add(&psig->ioac, &sig->ioac);
1141 write_sequnlock(&psig->stats_lock);
1142 spin_unlock_irq(¤t->sighand->siglock);
1143 }
1144
1145 retval = wo->wo_rusage
1146 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1147 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1148 ? p->signal->group_exit_code : p->exit_code;
1149 if (!retval && wo->wo_stat)
1150 retval = put_user(status, wo->wo_stat);
1151
1152 infop = wo->wo_info;
1153 if (!retval && infop)
1154 retval = put_user(SIGCHLD, &infop->si_signo);
1155 if (!retval && infop)
1156 retval = put_user(0, &infop->si_errno);
1157 if (!retval && infop) {
1158 int why;
1159
1160 if ((status & 0x7f) == 0) {
1161 why = CLD_EXITED;
1162 status >>= 8;
1163 } else {
1164 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1165 status &= 0x7f;
1166 }
1167 retval = put_user((short)why, &infop->si_code);
1168 if (!retval)
1169 retval = put_user(status, &infop->si_status);
1170 }
1171 if (!retval && infop)
1172 retval = put_user(pid, &infop->si_pid);
1173 if (!retval && infop)
1174 retval = put_user(uid, &infop->si_uid);
1175 if (!retval)
1176 retval = pid;
1177
1178 if (state == EXIT_TRACE) {
1179 write_lock_irq(&tasklist_lock);
1180 /* We dropped tasklist, ptracer could die and untrace */
1181 ptrace_unlink(p);
1182
1183 /* If parent wants a zombie, don't release it now */
1184 state = EXIT_ZOMBIE;
1185 if (do_notify_parent(p, p->exit_signal))
1186 state = EXIT_DEAD;
1187 p->exit_state = state;
1188 write_unlock_irq(&tasklist_lock);
1189 }
1190 if (state == EXIT_DEAD)
1191 release_task(p);
1192
1193 return retval;
1194}
1195
1196static int *task_stopped_code(struct task_struct *p, bool ptrace)
1197{
1198 if (ptrace) {
1199 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1200 return &p->exit_code;
1201 } else {
1202 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1203 return &p->signal->group_exit_code;
1204 }
1205 return NULL;
1206}
1207
1208/**
1209 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1210 * @wo: wait options
1211 * @ptrace: is the wait for ptrace
1212 * @p: task to wait for
1213 *
1214 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1215 *
1216 * CONTEXT:
1217 * read_lock(&tasklist_lock), which is released if return value is
1218 * non-zero. Also, grabs and releases @p->sighand->siglock.
1219 *
1220 * RETURNS:
1221 * 0 if wait condition didn't exist and search for other wait conditions
1222 * should continue. Non-zero return, -errno on failure and @p's pid on
1223 * success, implies that tasklist_lock is released and wait condition
1224 * search should terminate.
1225 */
1226static int wait_task_stopped(struct wait_opts *wo,
1227 int ptrace, struct task_struct *p)
1228{
1229 struct siginfo __user *infop;
1230 int retval, exit_code, *p_code, why;
1231 uid_t uid = 0; /* unneeded, required by compiler */
1232 pid_t pid;
1233
1234 /*
1235 * Traditionally we see ptrace'd stopped tasks regardless of options.
1236 */
1237 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1238 return 0;
1239
1240 if (!task_stopped_code(p, ptrace))
1241 return 0;
1242
1243 exit_code = 0;
1244 spin_lock_irq(&p->sighand->siglock);
1245
1246 p_code = task_stopped_code(p, ptrace);
1247 if (unlikely(!p_code))
1248 goto unlock_sig;
1249
1250 exit_code = *p_code;
1251 if (!exit_code)
1252 goto unlock_sig;
1253
1254 if (!unlikely(wo->wo_flags & WNOWAIT))
1255 *p_code = 0;
1256
1257 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1258unlock_sig:
1259 spin_unlock_irq(&p->sighand->siglock);
1260 if (!exit_code)
1261 return 0;
1262
1263 /*
1264 * Now we are pretty sure this task is interesting.
1265 * Make sure it doesn't get reaped out from under us while we
1266 * give up the lock and then examine it below. We don't want to
1267 * keep holding onto the tasklist_lock while we call getrusage and
1268 * possibly take page faults for user memory.
1269 */
1270 get_task_struct(p);
1271 pid = task_pid_vnr(p);
1272 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1273 read_unlock(&tasklist_lock);
1274 sched_annotate_sleep();
1275
1276 if (unlikely(wo->wo_flags & WNOWAIT))
1277 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1278
1279 retval = wo->wo_rusage
1280 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1281 if (!retval && wo->wo_stat)
1282 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1283
1284 infop = wo->wo_info;
1285 if (!retval && infop)
1286 retval = put_user(SIGCHLD, &infop->si_signo);
1287 if (!retval && infop)
1288 retval = put_user(0, &infop->si_errno);
1289 if (!retval && infop)
1290 retval = put_user((short)why, &infop->si_code);
1291 if (!retval && infop)
1292 retval = put_user(exit_code, &infop->si_status);
1293 if (!retval && infop)
1294 retval = put_user(pid, &infop->si_pid);
1295 if (!retval && infop)
1296 retval = put_user(uid, &infop->si_uid);
1297 if (!retval)
1298 retval = pid;
1299 put_task_struct(p);
1300
1301 BUG_ON(!retval);
1302 return retval;
1303}
1304
1305/*
1306 * Handle do_wait work for one task in a live, non-stopped state.
1307 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1308 * the lock and this task is uninteresting. If we return nonzero, we have
1309 * released the lock and the system call should return.
1310 */
1311static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1312{
1313 int retval;
1314 pid_t pid;
1315 uid_t uid;
1316
1317 if (!unlikely(wo->wo_flags & WCONTINUED))
1318 return 0;
1319
1320 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1321 return 0;
1322
1323 spin_lock_irq(&p->sighand->siglock);
1324 /* Re-check with the lock held. */
1325 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1326 spin_unlock_irq(&p->sighand->siglock);
1327 return 0;
1328 }
1329 if (!unlikely(wo->wo_flags & WNOWAIT))
1330 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1331 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1332 spin_unlock_irq(&p->sighand->siglock);
1333
1334 pid = task_pid_vnr(p);
1335 get_task_struct(p);
1336 read_unlock(&tasklist_lock);
1337 sched_annotate_sleep();
1338
1339 if (!wo->wo_info) {
1340 retval = wo->wo_rusage
1341 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1342 put_task_struct(p);
1343 if (!retval && wo->wo_stat)
1344 retval = put_user(0xffff, wo->wo_stat);
1345 if (!retval)
1346 retval = pid;
1347 } else {
1348 retval = wait_noreap_copyout(wo, p, pid, uid,
1349 CLD_CONTINUED, SIGCONT);
1350 BUG_ON(retval == 0);
1351 }
1352
1353 return retval;
1354}
1355
1356/*
1357 * Consider @p for a wait by @parent.
1358 *
1359 * -ECHILD should be in ->notask_error before the first call.
1360 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1361 * Returns zero if the search for a child should continue;
1362 * then ->notask_error is 0 if @p is an eligible child,
1363 * or another error from security_task_wait(), or still -ECHILD.
1364 */
1365static int wait_consider_task(struct wait_opts *wo, int ptrace,
1366 struct task_struct *p)
1367{
1368 /*
1369 * We can race with wait_task_zombie() from another thread.
1370 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1371 * can't confuse the checks below.
1372 */
1373 int exit_state = ACCESS_ONCE(p->exit_state);
1374 int ret;
1375
1376 if (unlikely(exit_state == EXIT_DEAD))
1377 return 0;
1378
1379 ret = eligible_child(wo, ptrace, p);
1380 if (!ret)
1381 return ret;
1382
1383 ret = security_task_wait(p);
1384 if (unlikely(ret < 0)) {
1385 /*
1386 * If we have not yet seen any eligible child,
1387 * then let this error code replace -ECHILD.
1388 * A permission error will give the user a clue
1389 * to look for security policy problems, rather
1390 * than for mysterious wait bugs.
1391 */
1392 if (wo->notask_error)
1393 wo->notask_error = ret;
1394 return 0;
1395 }
1396
1397 if (unlikely(exit_state == EXIT_TRACE)) {
1398 /*
1399 * ptrace == 0 means we are the natural parent. In this case
1400 * we should clear notask_error, debugger will notify us.
1401 */
1402 if (likely(!ptrace))
1403 wo->notask_error = 0;
1404 return 0;
1405 }
1406
1407 if (likely(!ptrace) && unlikely(p->ptrace)) {
1408 /*
1409 * If it is traced by its real parent's group, just pretend
1410 * the caller is ptrace_do_wait() and reap this child if it
1411 * is zombie.
1412 *
1413 * This also hides group stop state from real parent; otherwise
1414 * a single stop can be reported twice as group and ptrace stop.
1415 * If a ptracer wants to distinguish these two events for its
1416 * own children it should create a separate process which takes
1417 * the role of real parent.
1418 */
1419 if (!ptrace_reparented(p))
1420 ptrace = 1;
1421 }
1422
1423 /* slay zombie? */
1424 if (exit_state == EXIT_ZOMBIE) {
1425 /* we don't reap group leaders with subthreads */
1426 if (!delay_group_leader(p)) {
1427 /*
1428 * A zombie ptracee is only visible to its ptracer.
1429 * Notification and reaping will be cascaded to the
1430 * real parent when the ptracer detaches.
1431 */
1432 if (unlikely(ptrace) || likely(!p->ptrace))
1433 return wait_task_zombie(wo, p);
1434 }
1435
1436 /*
1437 * Allow access to stopped/continued state via zombie by
1438 * falling through. Clearing of notask_error is complex.
1439 *
1440 * When !@ptrace:
1441 *
1442 * If WEXITED is set, notask_error should naturally be
1443 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1444 * so, if there are live subthreads, there are events to
1445 * wait for. If all subthreads are dead, it's still safe
1446 * to clear - this function will be called again in finite
1447 * amount time once all the subthreads are released and
1448 * will then return without clearing.
1449 *
1450 * When @ptrace:
1451 *
1452 * Stopped state is per-task and thus can't change once the
1453 * target task dies. Only continued and exited can happen.
1454 * Clear notask_error if WCONTINUED | WEXITED.
1455 */
1456 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1457 wo->notask_error = 0;
1458 } else {
1459 /*
1460 * @p is alive and it's gonna stop, continue or exit, so
1461 * there always is something to wait for.
1462 */
1463 wo->notask_error = 0;
1464 }
1465
1466 /*
1467 * Wait for stopped. Depending on @ptrace, different stopped state
1468 * is used and the two don't interact with each other.
1469 */
1470 ret = wait_task_stopped(wo, ptrace, p);
1471 if (ret)
1472 return ret;
1473
1474 /*
1475 * Wait for continued. There's only one continued state and the
1476 * ptracer can consume it which can confuse the real parent. Don't
1477 * use WCONTINUED from ptracer. You don't need or want it.
1478 */
1479 return wait_task_continued(wo, p);
1480}
1481
1482/*
1483 * Do the work of do_wait() for one thread in the group, @tsk.
1484 *
1485 * -ECHILD should be in ->notask_error before the first call.
1486 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1487 * Returns zero if the search for a child should continue; then
1488 * ->notask_error is 0 if there were any eligible children,
1489 * or another error from security_task_wait(), or still -ECHILD.
1490 */
1491static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1492{
1493 struct task_struct *p;
1494
1495 list_for_each_entry(p, &tsk->children, sibling) {
1496 int ret = wait_consider_task(wo, 0, p);
1497
1498 if (ret)
1499 return ret;
1500 }
1501
1502 return 0;
1503}
1504
1505static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1506{
1507 struct task_struct *p;
1508
1509 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1510 int ret = wait_consider_task(wo, 1, p);
1511
1512 if (ret)
1513 return ret;
1514 }
1515
1516 return 0;
1517}
1518
1519static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1520 int sync, void *key)
1521{
1522 struct wait_opts *wo = container_of(wait, struct wait_opts,
1523 child_wait);
1524 struct task_struct *p = key;
1525
1526 if (!eligible_pid(wo, p))
1527 return 0;
1528
1529 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1530 return 0;
1531
1532 return default_wake_function(wait, mode, sync, key);
1533}
1534
1535void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1536{
1537 __wake_up_sync_key(&parent->signal->wait_chldexit,
1538 TASK_INTERRUPTIBLE, 1, p);
1539}
1540
1541static long do_wait(struct wait_opts *wo)
1542{
1543 struct task_struct *tsk;
1544 int retval;
1545
1546 trace_sched_process_wait(wo->wo_pid);
1547
1548 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1549 wo->child_wait.private = current;
1550 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1551repeat:
1552 /*
1553 * If there is nothing that can match our criteria, just get out.
1554 * We will clear ->notask_error to zero if we see any child that
1555 * might later match our criteria, even if we are not able to reap
1556 * it yet.
1557 */
1558 wo->notask_error = -ECHILD;
1559 if ((wo->wo_type < PIDTYPE_MAX) &&
1560 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1561 goto notask;
1562
1563 set_current_state(TASK_INTERRUPTIBLE);
1564 read_lock(&tasklist_lock);
1565 tsk = current;
1566 do {
1567 retval = do_wait_thread(wo, tsk);
1568 if (retval)
1569 goto end;
1570
1571 retval = ptrace_do_wait(wo, tsk);
1572 if (retval)
1573 goto end;
1574
1575 if (wo->wo_flags & __WNOTHREAD)
1576 break;
1577 } while_each_thread(current, tsk);
1578 read_unlock(&tasklist_lock);
1579
1580notask:
1581 retval = wo->notask_error;
1582 if (!retval && !(wo->wo_flags & WNOHANG)) {
1583 retval = -ERESTARTSYS;
1584 if (!signal_pending(current)) {
1585 schedule();
1586 goto repeat;
1587 }
1588 }
1589end:
1590 __set_current_state(TASK_RUNNING);
1591 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1592 return retval;
1593}
1594
1595SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1596 infop, int, options, struct rusage __user *, ru)
1597{
1598 struct wait_opts wo;
1599 struct pid *pid = NULL;
1600 enum pid_type type;
1601 long ret;
1602
1603 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1604 __WNOTHREAD|__WCLONE|__WALL))
1605 return -EINVAL;
1606 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1607 return -EINVAL;
1608
1609 switch (which) {
1610 case P_ALL:
1611 type = PIDTYPE_MAX;
1612 break;
1613 case P_PID:
1614 type = PIDTYPE_PID;
1615 if (upid <= 0)
1616 return -EINVAL;
1617 break;
1618 case P_PGID:
1619 type = PIDTYPE_PGID;
1620 if (upid <= 0)
1621 return -EINVAL;
1622 break;
1623 default:
1624 return -EINVAL;
1625 }
1626
1627 if (type < PIDTYPE_MAX)
1628 pid = find_get_pid(upid);
1629
1630 wo.wo_type = type;
1631 wo.wo_pid = pid;
1632 wo.wo_flags = options;
1633 wo.wo_info = infop;
1634 wo.wo_stat = NULL;
1635 wo.wo_rusage = ru;
1636 ret = do_wait(&wo);
1637
1638 if (ret > 0) {
1639 ret = 0;
1640 } else if (infop) {
1641 /*
1642 * For a WNOHANG return, clear out all the fields
1643 * we would set so the user can easily tell the
1644 * difference.
1645 */
1646 if (!ret)
1647 ret = put_user(0, &infop->si_signo);
1648 if (!ret)
1649 ret = put_user(0, &infop->si_errno);
1650 if (!ret)
1651 ret = put_user(0, &infop->si_code);
1652 if (!ret)
1653 ret = put_user(0, &infop->si_pid);
1654 if (!ret)
1655 ret = put_user(0, &infop->si_uid);
1656 if (!ret)
1657 ret = put_user(0, &infop->si_status);
1658 }
1659
1660 put_pid(pid);
1661 return ret;
1662}
1663
1664SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1665 int, options, struct rusage __user *, ru)
1666{
1667 struct wait_opts wo;
1668 struct pid *pid = NULL;
1669 enum pid_type type;
1670 long ret;
1671
1672 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1673 __WNOTHREAD|__WCLONE|__WALL))
1674 return -EINVAL;
1675
1676 if (upid == -1)
1677 type = PIDTYPE_MAX;
1678 else if (upid < 0) {
1679 type = PIDTYPE_PGID;
1680 pid = find_get_pid(-upid);
1681 } else if (upid == 0) {
1682 type = PIDTYPE_PGID;
1683 pid = get_task_pid(current, PIDTYPE_PGID);
1684 } else /* upid > 0 */ {
1685 type = PIDTYPE_PID;
1686 pid = find_get_pid(upid);
1687 }
1688
1689 wo.wo_type = type;
1690 wo.wo_pid = pid;
1691 wo.wo_flags = options | WEXITED;
1692 wo.wo_info = NULL;
1693 wo.wo_stat = stat_addr;
1694 wo.wo_rusage = ru;
1695 ret = do_wait(&wo);
1696 put_pid(pid);
1697
1698 return ret;
1699}
1700
1701#ifdef __ARCH_WANT_SYS_WAITPID
1702
1703/*
1704 * sys_waitpid() remains for compatibility. waitpid() should be
1705 * implemented by calling sys_wait4() from libc.a.
1706 */
1707SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1708{
1709 return sys_wait4(pid, stat_addr, options, NULL);
1710}
1711
1712#endif
1/*
2 * linux/kernel/exit.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7#include <linux/mm.h>
8#include <linux/slab.h>
9#include <linux/interrupt.h>
10#include <linux/module.h>
11#include <linux/capability.h>
12#include <linux/completion.h>
13#include <linux/personality.h>
14#include <linux/tty.h>
15#include <linux/iocontext.h>
16#include <linux/key.h>
17#include <linux/security.h>
18#include <linux/cpu.h>
19#include <linux/acct.h>
20#include <linux/tsacct_kern.h>
21#include <linux/file.h>
22#include <linux/fdtable.h>
23#include <linux/binfmts.h>
24#include <linux/nsproxy.h>
25#include <linux/pid_namespace.h>
26#include <linux/ptrace.h>
27#include <linux/profile.h>
28#include <linux/mount.h>
29#include <linux/proc_fs.h>
30#include <linux/kthread.h>
31#include <linux/mempolicy.h>
32#include <linux/taskstats_kern.h>
33#include <linux/delayacct.h>
34#include <linux/freezer.h>
35#include <linux/cgroup.h>
36#include <linux/syscalls.h>
37#include <linux/signal.h>
38#include <linux/posix-timers.h>
39#include <linux/cn_proc.h>
40#include <linux/mutex.h>
41#include <linux/futex.h>
42#include <linux/pipe_fs_i.h>
43#include <linux/audit.h> /* for audit_free() */
44#include <linux/resource.h>
45#include <linux/blkdev.h>
46#include <linux/task_io_accounting_ops.h>
47#include <linux/tracehook.h>
48#include <linux/fs_struct.h>
49#include <linux/init_task.h>
50#include <linux/perf_event.h>
51#include <trace/events/sched.h>
52#include <linux/hw_breakpoint.h>
53#include <linux/oom.h>
54#include <linux/writeback.h>
55#include <linux/shm.h>
56
57#include <asm/uaccess.h>
58#include <asm/unistd.h>
59#include <asm/pgtable.h>
60#include <asm/mmu_context.h>
61
62static void exit_mm(struct task_struct * tsk);
63
64static void __unhash_process(struct task_struct *p, bool group_dead)
65{
66 nr_threads--;
67 detach_pid(p, PIDTYPE_PID);
68 if (group_dead) {
69 detach_pid(p, PIDTYPE_PGID);
70 detach_pid(p, PIDTYPE_SID);
71
72 list_del_rcu(&p->tasks);
73 list_del_init(&p->sibling);
74 __this_cpu_dec(process_counts);
75 /*
76 * If we are the last child process in a pid namespace to be
77 * reaped, notify the reaper sleeping zap_pid_ns_processes().
78 */
79 if (IS_ENABLED(CONFIG_PID_NS)) {
80 struct task_struct *parent = p->real_parent;
81
82 if ((task_active_pid_ns(parent)->child_reaper == parent) &&
83 list_empty(&parent->children) &&
84 (parent->flags & PF_EXITING))
85 wake_up_process(parent);
86 }
87 }
88 list_del_rcu(&p->thread_group);
89}
90
91/*
92 * This function expects the tasklist_lock write-locked.
93 */
94static void __exit_signal(struct task_struct *tsk)
95{
96 struct signal_struct *sig = tsk->signal;
97 bool group_dead = thread_group_leader(tsk);
98 struct sighand_struct *sighand;
99 struct tty_struct *uninitialized_var(tty);
100
101 sighand = rcu_dereference_check(tsk->sighand,
102 lockdep_tasklist_lock_is_held());
103 spin_lock(&sighand->siglock);
104
105 posix_cpu_timers_exit(tsk);
106 if (group_dead) {
107 posix_cpu_timers_exit_group(tsk);
108 tty = sig->tty;
109 sig->tty = NULL;
110 } else {
111 /*
112 * This can only happen if the caller is de_thread().
113 * FIXME: this is the temporary hack, we should teach
114 * posix-cpu-timers to handle this case correctly.
115 */
116 if (unlikely(has_group_leader_pid(tsk)))
117 posix_cpu_timers_exit_group(tsk);
118
119 /*
120 * If there is any task waiting for the group exit
121 * then notify it:
122 */
123 if (sig->notify_count > 0 && !--sig->notify_count)
124 wake_up_process(sig->group_exit_task);
125
126 if (tsk == sig->curr_target)
127 sig->curr_target = next_thread(tsk);
128 /*
129 * Accumulate here the counters for all threads but the
130 * group leader as they die, so they can be added into
131 * the process-wide totals when those are taken.
132 * The group leader stays around as a zombie as long
133 * as there are other threads. When it gets reaped,
134 * the exit.c code will add its counts into these totals.
135 * We won't ever get here for the group leader, since it
136 * will have been the last reference on the signal_struct.
137 */
138 sig->utime += tsk->utime;
139 sig->stime += tsk->stime;
140 sig->gtime += tsk->gtime;
141 sig->min_flt += tsk->min_flt;
142 sig->maj_flt += tsk->maj_flt;
143 sig->nvcsw += tsk->nvcsw;
144 sig->nivcsw += tsk->nivcsw;
145 sig->inblock += task_io_get_inblock(tsk);
146 sig->oublock += task_io_get_oublock(tsk);
147 task_io_accounting_add(&sig->ioac, &tsk->ioac);
148 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
149 }
150
151 sig->nr_threads--;
152 __unhash_process(tsk, group_dead);
153
154 /*
155 * Do this under ->siglock, we can race with another thread
156 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
157 */
158 flush_sigqueue(&tsk->pending);
159 tsk->sighand = NULL;
160 spin_unlock(&sighand->siglock);
161
162 __cleanup_sighand(sighand);
163 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
164 if (group_dead) {
165 flush_sigqueue(&sig->shared_pending);
166 tty_kref_put(tty);
167 }
168}
169
170static void delayed_put_task_struct(struct rcu_head *rhp)
171{
172 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
173
174 perf_event_delayed_put(tsk);
175 trace_sched_process_free(tsk);
176 put_task_struct(tsk);
177}
178
179
180void release_task(struct task_struct * p)
181{
182 struct task_struct *leader;
183 int zap_leader;
184repeat:
185 /* don't need to get the RCU readlock here - the process is dead and
186 * can't be modifying its own credentials. But shut RCU-lockdep up */
187 rcu_read_lock();
188 atomic_dec(&__task_cred(p)->user->processes);
189 rcu_read_unlock();
190
191 proc_flush_task(p);
192
193 write_lock_irq(&tasklist_lock);
194 ptrace_release_task(p);
195 __exit_signal(p);
196
197 /*
198 * If we are the last non-leader member of the thread
199 * group, and the leader is zombie, then notify the
200 * group leader's parent process. (if it wants notification.)
201 */
202 zap_leader = 0;
203 leader = p->group_leader;
204 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
205 /*
206 * If we were the last child thread and the leader has
207 * exited already, and the leader's parent ignores SIGCHLD,
208 * then we are the one who should release the leader.
209 */
210 zap_leader = do_notify_parent(leader, leader->exit_signal);
211 if (zap_leader)
212 leader->exit_state = EXIT_DEAD;
213 }
214
215 write_unlock_irq(&tasklist_lock);
216 release_thread(p);
217 call_rcu(&p->rcu, delayed_put_task_struct);
218
219 p = leader;
220 if (unlikely(zap_leader))
221 goto repeat;
222}
223
224/*
225 * This checks not only the pgrp, but falls back on the pid if no
226 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
227 * without this...
228 *
229 * The caller must hold rcu lock or the tasklist lock.
230 */
231struct pid *session_of_pgrp(struct pid *pgrp)
232{
233 struct task_struct *p;
234 struct pid *sid = NULL;
235
236 p = pid_task(pgrp, PIDTYPE_PGID);
237 if (p == NULL)
238 p = pid_task(pgrp, PIDTYPE_PID);
239 if (p != NULL)
240 sid = task_session(p);
241
242 return sid;
243}
244
245/*
246 * Determine if a process group is "orphaned", according to the POSIX
247 * definition in 2.2.2.52. Orphaned process groups are not to be affected
248 * by terminal-generated stop signals. Newly orphaned process groups are
249 * to receive a SIGHUP and a SIGCONT.
250 *
251 * "I ask you, have you ever known what it is to be an orphan?"
252 */
253static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
254{
255 struct task_struct *p;
256
257 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
258 if ((p == ignored_task) ||
259 (p->exit_state && thread_group_empty(p)) ||
260 is_global_init(p->real_parent))
261 continue;
262
263 if (task_pgrp(p->real_parent) != pgrp &&
264 task_session(p->real_parent) == task_session(p))
265 return 0;
266 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
267
268 return 1;
269}
270
271int is_current_pgrp_orphaned(void)
272{
273 int retval;
274
275 read_lock(&tasklist_lock);
276 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
277 read_unlock(&tasklist_lock);
278
279 return retval;
280}
281
282static bool has_stopped_jobs(struct pid *pgrp)
283{
284 struct task_struct *p;
285
286 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
287 if (p->signal->flags & SIGNAL_STOP_STOPPED)
288 return true;
289 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
290
291 return false;
292}
293
294/*
295 * Check to see if any process groups have become orphaned as
296 * a result of our exiting, and if they have any stopped jobs,
297 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
298 */
299static void
300kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
301{
302 struct pid *pgrp = task_pgrp(tsk);
303 struct task_struct *ignored_task = tsk;
304
305 if (!parent)
306 /* exit: our father is in a different pgrp than
307 * we are and we were the only connection outside.
308 */
309 parent = tsk->real_parent;
310 else
311 /* reparent: our child is in a different pgrp than
312 * we are, and it was the only connection outside.
313 */
314 ignored_task = NULL;
315
316 if (task_pgrp(parent) != pgrp &&
317 task_session(parent) == task_session(tsk) &&
318 will_become_orphaned_pgrp(pgrp, ignored_task) &&
319 has_stopped_jobs(pgrp)) {
320 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
321 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
322 }
323}
324
325/**
326 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
327 *
328 * If a kernel thread is launched as a result of a system call, or if
329 * it ever exits, it should generally reparent itself to kthreadd so it
330 * isn't in the way of other processes and is correctly cleaned up on exit.
331 *
332 * The various task state such as scheduling policy and priority may have
333 * been inherited from a user process, so we reset them to sane values here.
334 *
335 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
336 */
337static void reparent_to_kthreadd(void)
338{
339 write_lock_irq(&tasklist_lock);
340
341 ptrace_unlink(current);
342 /* Reparent to init */
343 current->real_parent = current->parent = kthreadd_task;
344 list_move_tail(¤t->sibling, ¤t->real_parent->children);
345
346 /* Set the exit signal to SIGCHLD so we signal init on exit */
347 current->exit_signal = SIGCHLD;
348
349 if (task_nice(current) < 0)
350 set_user_nice(current, 0);
351 /* cpus_allowed? */
352 /* rt_priority? */
353 /* signals? */
354 memcpy(current->signal->rlim, init_task.signal->rlim,
355 sizeof(current->signal->rlim));
356
357 atomic_inc(&init_cred.usage);
358 commit_creds(&init_cred);
359 write_unlock_irq(&tasklist_lock);
360}
361
362void __set_special_pids(struct pid *pid)
363{
364 struct task_struct *curr = current->group_leader;
365
366 if (task_session(curr) != pid)
367 change_pid(curr, PIDTYPE_SID, pid);
368
369 if (task_pgrp(curr) != pid)
370 change_pid(curr, PIDTYPE_PGID, pid);
371}
372
373static void set_special_pids(struct pid *pid)
374{
375 write_lock_irq(&tasklist_lock);
376 __set_special_pids(pid);
377 write_unlock_irq(&tasklist_lock);
378}
379
380/*
381 * Let kernel threads use this to say that they allow a certain signal.
382 * Must not be used if kthread was cloned with CLONE_SIGHAND.
383 */
384int allow_signal(int sig)
385{
386 if (!valid_signal(sig) || sig < 1)
387 return -EINVAL;
388
389 spin_lock_irq(¤t->sighand->siglock);
390 /* This is only needed for daemonize()'ed kthreads */
391 sigdelset(¤t->blocked, sig);
392 /*
393 * Kernel threads handle their own signals. Let the signal code
394 * know it'll be handled, so that they don't get converted to
395 * SIGKILL or just silently dropped.
396 */
397 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
398 recalc_sigpending();
399 spin_unlock_irq(¤t->sighand->siglock);
400 return 0;
401}
402
403EXPORT_SYMBOL(allow_signal);
404
405int disallow_signal(int sig)
406{
407 if (!valid_signal(sig) || sig < 1)
408 return -EINVAL;
409
410 spin_lock_irq(¤t->sighand->siglock);
411 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
412 recalc_sigpending();
413 spin_unlock_irq(¤t->sighand->siglock);
414 return 0;
415}
416
417EXPORT_SYMBOL(disallow_signal);
418
419/*
420 * Put all the gunge required to become a kernel thread without
421 * attached user resources in one place where it belongs.
422 */
423
424void daemonize(const char *name, ...)
425{
426 va_list args;
427 sigset_t blocked;
428
429 va_start(args, name);
430 vsnprintf(current->comm, sizeof(current->comm), name, args);
431 va_end(args);
432
433 /*
434 * If we were started as result of loading a module, close all of the
435 * user space pages. We don't need them, and if we didn't close them
436 * they would be locked into memory.
437 */
438 exit_mm(current);
439 /*
440 * We don't want to get frozen, in case system-wide hibernation
441 * or suspend transition begins right now.
442 */
443 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
444
445 if (current->nsproxy != &init_nsproxy) {
446 get_nsproxy(&init_nsproxy);
447 switch_task_namespaces(current, &init_nsproxy);
448 }
449 set_special_pids(&init_struct_pid);
450 proc_clear_tty(current);
451
452 /* Block and flush all signals */
453 sigfillset(&blocked);
454 sigprocmask(SIG_BLOCK, &blocked, NULL);
455 flush_signals(current);
456
457 /* Become as one with the init task */
458
459 daemonize_fs_struct();
460 exit_files(current);
461 current->files = init_task.files;
462 atomic_inc(¤t->files->count);
463
464 reparent_to_kthreadd();
465}
466
467EXPORT_SYMBOL(daemonize);
468
469static void close_files(struct files_struct * files)
470{
471 int i, j;
472 struct fdtable *fdt;
473
474 j = 0;
475
476 /*
477 * It is safe to dereference the fd table without RCU or
478 * ->file_lock because this is the last reference to the
479 * files structure. But use RCU to shut RCU-lockdep up.
480 */
481 rcu_read_lock();
482 fdt = files_fdtable(files);
483 rcu_read_unlock();
484 for (;;) {
485 unsigned long set;
486 i = j * BITS_PER_LONG;
487 if (i >= fdt->max_fds)
488 break;
489 set = fdt->open_fds[j++];
490 while (set) {
491 if (set & 1) {
492 struct file * file = xchg(&fdt->fd[i], NULL);
493 if (file) {
494 filp_close(file, files);
495 cond_resched();
496 }
497 }
498 i++;
499 set >>= 1;
500 }
501 }
502}
503
504struct files_struct *get_files_struct(struct task_struct *task)
505{
506 struct files_struct *files;
507
508 task_lock(task);
509 files = task->files;
510 if (files)
511 atomic_inc(&files->count);
512 task_unlock(task);
513
514 return files;
515}
516
517void put_files_struct(struct files_struct *files)
518{
519 struct fdtable *fdt;
520
521 if (atomic_dec_and_test(&files->count)) {
522 close_files(files);
523 /*
524 * Free the fd and fdset arrays if we expanded them.
525 * If the fdtable was embedded, pass files for freeing
526 * at the end of the RCU grace period. Otherwise,
527 * you can free files immediately.
528 */
529 rcu_read_lock();
530 fdt = files_fdtable(files);
531 if (fdt != &files->fdtab)
532 kmem_cache_free(files_cachep, files);
533 free_fdtable(fdt);
534 rcu_read_unlock();
535 }
536}
537
538void reset_files_struct(struct files_struct *files)
539{
540 struct task_struct *tsk = current;
541 struct files_struct *old;
542
543 old = tsk->files;
544 task_lock(tsk);
545 tsk->files = files;
546 task_unlock(tsk);
547 put_files_struct(old);
548}
549
550void exit_files(struct task_struct *tsk)
551{
552 struct files_struct * files = tsk->files;
553
554 if (files) {
555 task_lock(tsk);
556 tsk->files = NULL;
557 task_unlock(tsk);
558 put_files_struct(files);
559 }
560}
561
562#ifdef CONFIG_MM_OWNER
563/*
564 * A task is exiting. If it owned this mm, find a new owner for the mm.
565 */
566void mm_update_next_owner(struct mm_struct *mm)
567{
568 struct task_struct *c, *g, *p = current;
569
570retry:
571 /*
572 * If the exiting or execing task is not the owner, it's
573 * someone else's problem.
574 */
575 if (mm->owner != p)
576 return;
577 /*
578 * The current owner is exiting/execing and there are no other
579 * candidates. Do not leave the mm pointing to a possibly
580 * freed task structure.
581 */
582 if (atomic_read(&mm->mm_users) <= 1) {
583 mm->owner = NULL;
584 return;
585 }
586
587 read_lock(&tasklist_lock);
588 /*
589 * Search in the children
590 */
591 list_for_each_entry(c, &p->children, sibling) {
592 if (c->mm == mm)
593 goto assign_new_owner;
594 }
595
596 /*
597 * Search in the siblings
598 */
599 list_for_each_entry(c, &p->real_parent->children, sibling) {
600 if (c->mm == mm)
601 goto assign_new_owner;
602 }
603
604 /*
605 * Search through everything else. We should not get
606 * here often
607 */
608 do_each_thread(g, c) {
609 if (c->mm == mm)
610 goto assign_new_owner;
611 } while_each_thread(g, c);
612
613 read_unlock(&tasklist_lock);
614 /*
615 * We found no owner yet mm_users > 1: this implies that we are
616 * most likely racing with swapoff (try_to_unuse()) or /proc or
617 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
618 */
619 mm->owner = NULL;
620 return;
621
622assign_new_owner:
623 BUG_ON(c == p);
624 get_task_struct(c);
625 /*
626 * The task_lock protects c->mm from changing.
627 * We always want mm->owner->mm == mm
628 */
629 task_lock(c);
630 /*
631 * Delay read_unlock() till we have the task_lock()
632 * to ensure that c does not slip away underneath us
633 */
634 read_unlock(&tasklist_lock);
635 if (c->mm != mm) {
636 task_unlock(c);
637 put_task_struct(c);
638 goto retry;
639 }
640 mm->owner = c;
641 task_unlock(c);
642 put_task_struct(c);
643}
644#endif /* CONFIG_MM_OWNER */
645
646/*
647 * Turn us into a lazy TLB process if we
648 * aren't already..
649 */
650static void exit_mm(struct task_struct * tsk)
651{
652 struct mm_struct *mm = tsk->mm;
653 struct core_state *core_state;
654
655 mm_release(tsk, mm);
656 if (!mm)
657 return;
658 sync_mm_rss(mm);
659 /*
660 * Serialize with any possible pending coredump.
661 * We must hold mmap_sem around checking core_state
662 * and clearing tsk->mm. The core-inducing thread
663 * will increment ->nr_threads for each thread in the
664 * group with ->mm != NULL.
665 */
666 down_read(&mm->mmap_sem);
667 core_state = mm->core_state;
668 if (core_state) {
669 struct core_thread self;
670 up_read(&mm->mmap_sem);
671
672 self.task = tsk;
673 self.next = xchg(&core_state->dumper.next, &self);
674 /*
675 * Implies mb(), the result of xchg() must be visible
676 * to core_state->dumper.
677 */
678 if (atomic_dec_and_test(&core_state->nr_threads))
679 complete(&core_state->startup);
680
681 for (;;) {
682 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
683 if (!self.task) /* see coredump_finish() */
684 break;
685 schedule();
686 }
687 __set_task_state(tsk, TASK_RUNNING);
688 down_read(&mm->mmap_sem);
689 }
690 atomic_inc(&mm->mm_count);
691 BUG_ON(mm != tsk->active_mm);
692 /* more a memory barrier than a real lock */
693 task_lock(tsk);
694 tsk->mm = NULL;
695 up_read(&mm->mmap_sem);
696 enter_lazy_tlb(mm, current);
697 task_unlock(tsk);
698 mm_update_next_owner(mm);
699 mmput(mm);
700}
701
702/*
703 * When we die, we re-parent all our children, and try to:
704 * 1. give them to another thread in our thread group, if such a member exists
705 * 2. give it to the first ancestor process which prctl'd itself as a
706 * child_subreaper for its children (like a service manager)
707 * 3. give it to the init process (PID 1) in our pid namespace
708 */
709static struct task_struct *find_new_reaper(struct task_struct *father)
710 __releases(&tasklist_lock)
711 __acquires(&tasklist_lock)
712{
713 struct pid_namespace *pid_ns = task_active_pid_ns(father);
714 struct task_struct *thread;
715
716 thread = father;
717 while_each_thread(father, thread) {
718 if (thread->flags & PF_EXITING)
719 continue;
720 if (unlikely(pid_ns->child_reaper == father))
721 pid_ns->child_reaper = thread;
722 return thread;
723 }
724
725 if (unlikely(pid_ns->child_reaper == father)) {
726 write_unlock_irq(&tasklist_lock);
727 if (unlikely(pid_ns == &init_pid_ns)) {
728 panic("Attempted to kill init! exitcode=0x%08x\n",
729 father->signal->group_exit_code ?:
730 father->exit_code);
731 }
732
733 zap_pid_ns_processes(pid_ns);
734 write_lock_irq(&tasklist_lock);
735 } else if (father->signal->has_child_subreaper) {
736 struct task_struct *reaper;
737
738 /*
739 * Find the first ancestor marked as child_subreaper.
740 * Note that the code below checks same_thread_group(reaper,
741 * pid_ns->child_reaper). This is what we need to DTRT in a
742 * PID namespace. However we still need the check above, see
743 * http://marc.info/?l=linux-kernel&m=131385460420380
744 */
745 for (reaper = father->real_parent;
746 reaper != &init_task;
747 reaper = reaper->real_parent) {
748 if (same_thread_group(reaper, pid_ns->child_reaper))
749 break;
750 if (!reaper->signal->is_child_subreaper)
751 continue;
752 thread = reaper;
753 do {
754 if (!(thread->flags & PF_EXITING))
755 return reaper;
756 } while_each_thread(reaper, thread);
757 }
758 }
759
760 return pid_ns->child_reaper;
761}
762
763/*
764* Any that need to be release_task'd are put on the @dead list.
765 */
766static void reparent_leader(struct task_struct *father, struct task_struct *p,
767 struct list_head *dead)
768{
769 list_move_tail(&p->sibling, &p->real_parent->children);
770
771 if (p->exit_state == EXIT_DEAD)
772 return;
773 /*
774 * If this is a threaded reparent there is no need to
775 * notify anyone anything has happened.
776 */
777 if (same_thread_group(p->real_parent, father))
778 return;
779
780 /* We don't want people slaying init. */
781 p->exit_signal = SIGCHLD;
782
783 /* If it has exited notify the new parent about this child's death. */
784 if (!p->ptrace &&
785 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
786 if (do_notify_parent(p, p->exit_signal)) {
787 p->exit_state = EXIT_DEAD;
788 list_move_tail(&p->sibling, dead);
789 }
790 }
791
792 kill_orphaned_pgrp(p, father);
793}
794
795static void forget_original_parent(struct task_struct *father)
796{
797 struct task_struct *p, *n, *reaper;
798 LIST_HEAD(dead_children);
799
800 write_lock_irq(&tasklist_lock);
801 /*
802 * Note that exit_ptrace() and find_new_reaper() might
803 * drop tasklist_lock and reacquire it.
804 */
805 exit_ptrace(father);
806 reaper = find_new_reaper(father);
807
808 list_for_each_entry_safe(p, n, &father->children, sibling) {
809 struct task_struct *t = p;
810 do {
811 t->real_parent = reaper;
812 if (t->parent == father) {
813 BUG_ON(t->ptrace);
814 t->parent = t->real_parent;
815 }
816 if (t->pdeath_signal)
817 group_send_sig_info(t->pdeath_signal,
818 SEND_SIG_NOINFO, t);
819 } while_each_thread(p, t);
820 reparent_leader(father, p, &dead_children);
821 }
822 write_unlock_irq(&tasklist_lock);
823
824 BUG_ON(!list_empty(&father->children));
825
826 list_for_each_entry_safe(p, n, &dead_children, sibling) {
827 list_del_init(&p->sibling);
828 release_task(p);
829 }
830}
831
832/*
833 * Send signals to all our closest relatives so that they know
834 * to properly mourn us..
835 */
836static void exit_notify(struct task_struct *tsk, int group_dead)
837{
838 bool autoreap;
839
840 /*
841 * This does two things:
842 *
843 * A. Make init inherit all the child processes
844 * B. Check to see if any process groups have become orphaned
845 * as a result of our exiting, and if they have any stopped
846 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
847 */
848 forget_original_parent(tsk);
849 exit_task_namespaces(tsk);
850
851 write_lock_irq(&tasklist_lock);
852 if (group_dead)
853 kill_orphaned_pgrp(tsk->group_leader, NULL);
854
855 if (unlikely(tsk->ptrace)) {
856 int sig = thread_group_leader(tsk) &&
857 thread_group_empty(tsk) &&
858 !ptrace_reparented(tsk) ?
859 tsk->exit_signal : SIGCHLD;
860 autoreap = do_notify_parent(tsk, sig);
861 } else if (thread_group_leader(tsk)) {
862 autoreap = thread_group_empty(tsk) &&
863 do_notify_parent(tsk, tsk->exit_signal);
864 } else {
865 autoreap = true;
866 }
867
868 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
869
870 /* mt-exec, de_thread() is waiting for group leader */
871 if (unlikely(tsk->signal->notify_count < 0))
872 wake_up_process(tsk->signal->group_exit_task);
873 write_unlock_irq(&tasklist_lock);
874
875 /* If the process is dead, release it - nobody will wait for it */
876 if (autoreap)
877 release_task(tsk);
878}
879
880#ifdef CONFIG_DEBUG_STACK_USAGE
881static void check_stack_usage(void)
882{
883 static DEFINE_SPINLOCK(low_water_lock);
884 static int lowest_to_date = THREAD_SIZE;
885 unsigned long free;
886
887 free = stack_not_used(current);
888
889 if (free >= lowest_to_date)
890 return;
891
892 spin_lock(&low_water_lock);
893 if (free < lowest_to_date) {
894 printk(KERN_WARNING "%s (%d) used greatest stack depth: "
895 "%lu bytes left\n",
896 current->comm, task_pid_nr(current), free);
897 lowest_to_date = free;
898 }
899 spin_unlock(&low_water_lock);
900}
901#else
902static inline void check_stack_usage(void) {}
903#endif
904
905void do_exit(long code)
906{
907 struct task_struct *tsk = current;
908 int group_dead;
909
910 profile_task_exit(tsk);
911
912 WARN_ON(blk_needs_flush_plug(tsk));
913
914 if (unlikely(in_interrupt()))
915 panic("Aiee, killing interrupt handler!");
916 if (unlikely(!tsk->pid))
917 panic("Attempted to kill the idle task!");
918
919 /*
920 * If do_exit is called because this processes oopsed, it's possible
921 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
922 * continuing. Amongst other possible reasons, this is to prevent
923 * mm_release()->clear_child_tid() from writing to a user-controlled
924 * kernel address.
925 */
926 set_fs(USER_DS);
927
928 ptrace_event(PTRACE_EVENT_EXIT, code);
929
930 validate_creds_for_do_exit(tsk);
931
932 /*
933 * We're taking recursive faults here in do_exit. Safest is to just
934 * leave this task alone and wait for reboot.
935 */
936 if (unlikely(tsk->flags & PF_EXITING)) {
937 printk(KERN_ALERT
938 "Fixing recursive fault but reboot is needed!\n");
939 /*
940 * We can do this unlocked here. The futex code uses
941 * this flag just to verify whether the pi state
942 * cleanup has been done or not. In the worst case it
943 * loops once more. We pretend that the cleanup was
944 * done as there is no way to return. Either the
945 * OWNER_DIED bit is set by now or we push the blocked
946 * task into the wait for ever nirwana as well.
947 */
948 tsk->flags |= PF_EXITPIDONE;
949 set_current_state(TASK_UNINTERRUPTIBLE);
950 schedule();
951 }
952
953 exit_signals(tsk); /* sets PF_EXITING */
954 /*
955 * tsk->flags are checked in the futex code to protect against
956 * an exiting task cleaning up the robust pi futexes, and in
957 * task_work_add() to avoid the race with exit_task_work().
958 */
959 smp_mb();
960 raw_spin_unlock_wait(&tsk->pi_lock);
961
962 exit_task_work(tsk);
963
964 if (unlikely(in_atomic()))
965 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
966 current->comm, task_pid_nr(current),
967 preempt_count());
968
969 acct_update_integrals(tsk);
970 /* sync mm's RSS info before statistics gathering */
971 if (tsk->mm)
972 sync_mm_rss(tsk->mm);
973 group_dead = atomic_dec_and_test(&tsk->signal->live);
974 if (group_dead) {
975 hrtimer_cancel(&tsk->signal->real_timer);
976 exit_itimers(tsk->signal);
977 if (tsk->mm)
978 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
979 }
980 acct_collect(code, group_dead);
981 if (group_dead)
982 tty_audit_exit();
983 audit_free(tsk);
984
985 tsk->exit_code = code;
986 taskstats_exit(tsk, group_dead);
987
988 exit_mm(tsk);
989
990 if (group_dead)
991 acct_process();
992 trace_sched_process_exit(tsk);
993
994 exit_sem(tsk);
995 exit_shm(tsk);
996 exit_files(tsk);
997 exit_fs(tsk);
998 check_stack_usage();
999 exit_thread();
1000
1001 /*
1002 * Flush inherited counters to the parent - before the parent
1003 * gets woken up by child-exit notifications.
1004 *
1005 * because of cgroup mode, must be called before cgroup_exit()
1006 */
1007 perf_event_exit_task(tsk);
1008
1009 cgroup_exit(tsk, 1);
1010
1011 if (group_dead)
1012 disassociate_ctty(1);
1013
1014 module_put(task_thread_info(tsk)->exec_domain->module);
1015
1016 proc_exit_connector(tsk);
1017
1018 /*
1019 * FIXME: do that only when needed, using sched_exit tracepoint
1020 */
1021 ptrace_put_breakpoints(tsk);
1022
1023 exit_notify(tsk, group_dead);
1024#ifdef CONFIG_NUMA
1025 task_lock(tsk);
1026 mpol_put(tsk->mempolicy);
1027 tsk->mempolicy = NULL;
1028 task_unlock(tsk);
1029#endif
1030#ifdef CONFIG_FUTEX
1031 if (unlikely(current->pi_state_cache))
1032 kfree(current->pi_state_cache);
1033#endif
1034 /*
1035 * Make sure we are holding no locks:
1036 */
1037 debug_check_no_locks_held(tsk);
1038 /*
1039 * We can do this unlocked here. The futex code uses this flag
1040 * just to verify whether the pi state cleanup has been done
1041 * or not. In the worst case it loops once more.
1042 */
1043 tsk->flags |= PF_EXITPIDONE;
1044
1045 if (tsk->io_context)
1046 exit_io_context(tsk);
1047
1048 if (tsk->splice_pipe)
1049 __free_pipe_info(tsk->splice_pipe);
1050
1051 validate_creds_for_do_exit(tsk);
1052
1053 preempt_disable();
1054 if (tsk->nr_dirtied)
1055 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
1056 exit_rcu();
1057
1058 /*
1059 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
1060 * when the following two conditions become true.
1061 * - There is race condition of mmap_sem (It is acquired by
1062 * exit_mm()), and
1063 * - SMI occurs before setting TASK_RUNINNG.
1064 * (or hypervisor of virtual machine switches to other guest)
1065 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
1066 *
1067 * To avoid it, we have to wait for releasing tsk->pi_lock which
1068 * is held by try_to_wake_up()
1069 */
1070 smp_mb();
1071 raw_spin_unlock_wait(&tsk->pi_lock);
1072
1073 /* causes final put_task_struct in finish_task_switch(). */
1074 tsk->state = TASK_DEAD;
1075 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
1076 schedule();
1077 BUG();
1078 /* Avoid "noreturn function does return". */
1079 for (;;)
1080 cpu_relax(); /* For when BUG is null */
1081}
1082
1083EXPORT_SYMBOL_GPL(do_exit);
1084
1085void complete_and_exit(struct completion *comp, long code)
1086{
1087 if (comp)
1088 complete(comp);
1089
1090 do_exit(code);
1091}
1092
1093EXPORT_SYMBOL(complete_and_exit);
1094
1095SYSCALL_DEFINE1(exit, int, error_code)
1096{
1097 do_exit((error_code&0xff)<<8);
1098}
1099
1100/*
1101 * Take down every thread in the group. This is called by fatal signals
1102 * as well as by sys_exit_group (below).
1103 */
1104void
1105do_group_exit(int exit_code)
1106{
1107 struct signal_struct *sig = current->signal;
1108
1109 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1110
1111 if (signal_group_exit(sig))
1112 exit_code = sig->group_exit_code;
1113 else if (!thread_group_empty(current)) {
1114 struct sighand_struct *const sighand = current->sighand;
1115 spin_lock_irq(&sighand->siglock);
1116 if (signal_group_exit(sig))
1117 /* Another thread got here before we took the lock. */
1118 exit_code = sig->group_exit_code;
1119 else {
1120 sig->group_exit_code = exit_code;
1121 sig->flags = SIGNAL_GROUP_EXIT;
1122 zap_other_threads(current);
1123 }
1124 spin_unlock_irq(&sighand->siglock);
1125 }
1126
1127 do_exit(exit_code);
1128 /* NOTREACHED */
1129}
1130
1131/*
1132 * this kills every thread in the thread group. Note that any externally
1133 * wait4()-ing process will get the correct exit code - even if this
1134 * thread is not the thread group leader.
1135 */
1136SYSCALL_DEFINE1(exit_group, int, error_code)
1137{
1138 do_group_exit((error_code & 0xff) << 8);
1139 /* NOTREACHED */
1140 return 0;
1141}
1142
1143struct wait_opts {
1144 enum pid_type wo_type;
1145 int wo_flags;
1146 struct pid *wo_pid;
1147
1148 struct siginfo __user *wo_info;
1149 int __user *wo_stat;
1150 struct rusage __user *wo_rusage;
1151
1152 wait_queue_t child_wait;
1153 int notask_error;
1154};
1155
1156static inline
1157struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1158{
1159 if (type != PIDTYPE_PID)
1160 task = task->group_leader;
1161 return task->pids[type].pid;
1162}
1163
1164static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1165{
1166 return wo->wo_type == PIDTYPE_MAX ||
1167 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1168}
1169
1170static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1171{
1172 if (!eligible_pid(wo, p))
1173 return 0;
1174 /* Wait for all children (clone and not) if __WALL is set;
1175 * otherwise, wait for clone children *only* if __WCLONE is
1176 * set; otherwise, wait for non-clone children *only*. (Note:
1177 * A "clone" child here is one that reports to its parent
1178 * using a signal other than SIGCHLD.) */
1179 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1180 && !(wo->wo_flags & __WALL))
1181 return 0;
1182
1183 return 1;
1184}
1185
1186static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1187 pid_t pid, uid_t uid, int why, int status)
1188{
1189 struct siginfo __user *infop;
1190 int retval = wo->wo_rusage
1191 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1192
1193 put_task_struct(p);
1194 infop = wo->wo_info;
1195 if (infop) {
1196 if (!retval)
1197 retval = put_user(SIGCHLD, &infop->si_signo);
1198 if (!retval)
1199 retval = put_user(0, &infop->si_errno);
1200 if (!retval)
1201 retval = put_user((short)why, &infop->si_code);
1202 if (!retval)
1203 retval = put_user(pid, &infop->si_pid);
1204 if (!retval)
1205 retval = put_user(uid, &infop->si_uid);
1206 if (!retval)
1207 retval = put_user(status, &infop->si_status);
1208 }
1209 if (!retval)
1210 retval = pid;
1211 return retval;
1212}
1213
1214/*
1215 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1216 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1217 * the lock and this task is uninteresting. If we return nonzero, we have
1218 * released the lock and the system call should return.
1219 */
1220static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1221{
1222 unsigned long state;
1223 int retval, status, traced;
1224 pid_t pid = task_pid_vnr(p);
1225 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1226 struct siginfo __user *infop;
1227
1228 if (!likely(wo->wo_flags & WEXITED))
1229 return 0;
1230
1231 if (unlikely(wo->wo_flags & WNOWAIT)) {
1232 int exit_code = p->exit_code;
1233 int why;
1234
1235 get_task_struct(p);
1236 read_unlock(&tasklist_lock);
1237 if ((exit_code & 0x7f) == 0) {
1238 why = CLD_EXITED;
1239 status = exit_code >> 8;
1240 } else {
1241 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1242 status = exit_code & 0x7f;
1243 }
1244 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1245 }
1246
1247 /*
1248 * Try to move the task's state to DEAD
1249 * only one thread is allowed to do this:
1250 */
1251 state = xchg(&p->exit_state, EXIT_DEAD);
1252 if (state != EXIT_ZOMBIE) {
1253 BUG_ON(state != EXIT_DEAD);
1254 return 0;
1255 }
1256
1257 traced = ptrace_reparented(p);
1258 /*
1259 * It can be ptraced but not reparented, check
1260 * thread_group_leader() to filter out sub-threads.
1261 */
1262 if (likely(!traced) && thread_group_leader(p)) {
1263 struct signal_struct *psig;
1264 struct signal_struct *sig;
1265 unsigned long maxrss;
1266 cputime_t tgutime, tgstime;
1267
1268 /*
1269 * The resource counters for the group leader are in its
1270 * own task_struct. Those for dead threads in the group
1271 * are in its signal_struct, as are those for the child
1272 * processes it has previously reaped. All these
1273 * accumulate in the parent's signal_struct c* fields.
1274 *
1275 * We don't bother to take a lock here to protect these
1276 * p->signal fields, because they are only touched by
1277 * __exit_signal, which runs with tasklist_lock
1278 * write-locked anyway, and so is excluded here. We do
1279 * need to protect the access to parent->signal fields,
1280 * as other threads in the parent group can be right
1281 * here reaping other children at the same time.
1282 *
1283 * We use thread_group_times() to get times for the thread
1284 * group, which consolidates times for all threads in the
1285 * group including the group leader.
1286 */
1287 thread_group_times(p, &tgutime, &tgstime);
1288 spin_lock_irq(&p->real_parent->sighand->siglock);
1289 psig = p->real_parent->signal;
1290 sig = p->signal;
1291 psig->cutime += tgutime + sig->cutime;
1292 psig->cstime += tgstime + sig->cstime;
1293 psig->cgtime += p->gtime + sig->gtime + sig->cgtime;
1294 psig->cmin_flt +=
1295 p->min_flt + sig->min_flt + sig->cmin_flt;
1296 psig->cmaj_flt +=
1297 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1298 psig->cnvcsw +=
1299 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1300 psig->cnivcsw +=
1301 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1302 psig->cinblock +=
1303 task_io_get_inblock(p) +
1304 sig->inblock + sig->cinblock;
1305 psig->coublock +=
1306 task_io_get_oublock(p) +
1307 sig->oublock + sig->coublock;
1308 maxrss = max(sig->maxrss, sig->cmaxrss);
1309 if (psig->cmaxrss < maxrss)
1310 psig->cmaxrss = maxrss;
1311 task_io_accounting_add(&psig->ioac, &p->ioac);
1312 task_io_accounting_add(&psig->ioac, &sig->ioac);
1313 spin_unlock_irq(&p->real_parent->sighand->siglock);
1314 }
1315
1316 /*
1317 * Now we are sure this task is interesting, and no other
1318 * thread can reap it because we set its state to EXIT_DEAD.
1319 */
1320 read_unlock(&tasklist_lock);
1321
1322 retval = wo->wo_rusage
1323 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1324 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1325 ? p->signal->group_exit_code : p->exit_code;
1326 if (!retval && wo->wo_stat)
1327 retval = put_user(status, wo->wo_stat);
1328
1329 infop = wo->wo_info;
1330 if (!retval && infop)
1331 retval = put_user(SIGCHLD, &infop->si_signo);
1332 if (!retval && infop)
1333 retval = put_user(0, &infop->si_errno);
1334 if (!retval && infop) {
1335 int why;
1336
1337 if ((status & 0x7f) == 0) {
1338 why = CLD_EXITED;
1339 status >>= 8;
1340 } else {
1341 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1342 status &= 0x7f;
1343 }
1344 retval = put_user((short)why, &infop->si_code);
1345 if (!retval)
1346 retval = put_user(status, &infop->si_status);
1347 }
1348 if (!retval && infop)
1349 retval = put_user(pid, &infop->si_pid);
1350 if (!retval && infop)
1351 retval = put_user(uid, &infop->si_uid);
1352 if (!retval)
1353 retval = pid;
1354
1355 if (traced) {
1356 write_lock_irq(&tasklist_lock);
1357 /* We dropped tasklist, ptracer could die and untrace */
1358 ptrace_unlink(p);
1359 /*
1360 * If this is not a sub-thread, notify the parent.
1361 * If parent wants a zombie, don't release it now.
1362 */
1363 if (thread_group_leader(p) &&
1364 !do_notify_parent(p, p->exit_signal)) {
1365 p->exit_state = EXIT_ZOMBIE;
1366 p = NULL;
1367 }
1368 write_unlock_irq(&tasklist_lock);
1369 }
1370 if (p != NULL)
1371 release_task(p);
1372
1373 return retval;
1374}
1375
1376static int *task_stopped_code(struct task_struct *p, bool ptrace)
1377{
1378 if (ptrace) {
1379 if (task_is_stopped_or_traced(p) &&
1380 !(p->jobctl & JOBCTL_LISTENING))
1381 return &p->exit_code;
1382 } else {
1383 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1384 return &p->signal->group_exit_code;
1385 }
1386 return NULL;
1387}
1388
1389/**
1390 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1391 * @wo: wait options
1392 * @ptrace: is the wait for ptrace
1393 * @p: task to wait for
1394 *
1395 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1396 *
1397 * CONTEXT:
1398 * read_lock(&tasklist_lock), which is released if return value is
1399 * non-zero. Also, grabs and releases @p->sighand->siglock.
1400 *
1401 * RETURNS:
1402 * 0 if wait condition didn't exist and search for other wait conditions
1403 * should continue. Non-zero return, -errno on failure and @p's pid on
1404 * success, implies that tasklist_lock is released and wait condition
1405 * search should terminate.
1406 */
1407static int wait_task_stopped(struct wait_opts *wo,
1408 int ptrace, struct task_struct *p)
1409{
1410 struct siginfo __user *infop;
1411 int retval, exit_code, *p_code, why;
1412 uid_t uid = 0; /* unneeded, required by compiler */
1413 pid_t pid;
1414
1415 /*
1416 * Traditionally we see ptrace'd stopped tasks regardless of options.
1417 */
1418 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1419 return 0;
1420
1421 if (!task_stopped_code(p, ptrace))
1422 return 0;
1423
1424 exit_code = 0;
1425 spin_lock_irq(&p->sighand->siglock);
1426
1427 p_code = task_stopped_code(p, ptrace);
1428 if (unlikely(!p_code))
1429 goto unlock_sig;
1430
1431 exit_code = *p_code;
1432 if (!exit_code)
1433 goto unlock_sig;
1434
1435 if (!unlikely(wo->wo_flags & WNOWAIT))
1436 *p_code = 0;
1437
1438 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1439unlock_sig:
1440 spin_unlock_irq(&p->sighand->siglock);
1441 if (!exit_code)
1442 return 0;
1443
1444 /*
1445 * Now we are pretty sure this task is interesting.
1446 * Make sure it doesn't get reaped out from under us while we
1447 * give up the lock and then examine it below. We don't want to
1448 * keep holding onto the tasklist_lock while we call getrusage and
1449 * possibly take page faults for user memory.
1450 */
1451 get_task_struct(p);
1452 pid = task_pid_vnr(p);
1453 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1454 read_unlock(&tasklist_lock);
1455
1456 if (unlikely(wo->wo_flags & WNOWAIT))
1457 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1458
1459 retval = wo->wo_rusage
1460 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1461 if (!retval && wo->wo_stat)
1462 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1463
1464 infop = wo->wo_info;
1465 if (!retval && infop)
1466 retval = put_user(SIGCHLD, &infop->si_signo);
1467 if (!retval && infop)
1468 retval = put_user(0, &infop->si_errno);
1469 if (!retval && infop)
1470 retval = put_user((short)why, &infop->si_code);
1471 if (!retval && infop)
1472 retval = put_user(exit_code, &infop->si_status);
1473 if (!retval && infop)
1474 retval = put_user(pid, &infop->si_pid);
1475 if (!retval && infop)
1476 retval = put_user(uid, &infop->si_uid);
1477 if (!retval)
1478 retval = pid;
1479 put_task_struct(p);
1480
1481 BUG_ON(!retval);
1482 return retval;
1483}
1484
1485/*
1486 * Handle do_wait work for one task in a live, non-stopped state.
1487 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1488 * the lock and this task is uninteresting. If we return nonzero, we have
1489 * released the lock and the system call should return.
1490 */
1491static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1492{
1493 int retval;
1494 pid_t pid;
1495 uid_t uid;
1496
1497 if (!unlikely(wo->wo_flags & WCONTINUED))
1498 return 0;
1499
1500 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1501 return 0;
1502
1503 spin_lock_irq(&p->sighand->siglock);
1504 /* Re-check with the lock held. */
1505 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1506 spin_unlock_irq(&p->sighand->siglock);
1507 return 0;
1508 }
1509 if (!unlikely(wo->wo_flags & WNOWAIT))
1510 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1511 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1512 spin_unlock_irq(&p->sighand->siglock);
1513
1514 pid = task_pid_vnr(p);
1515 get_task_struct(p);
1516 read_unlock(&tasklist_lock);
1517
1518 if (!wo->wo_info) {
1519 retval = wo->wo_rusage
1520 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1521 put_task_struct(p);
1522 if (!retval && wo->wo_stat)
1523 retval = put_user(0xffff, wo->wo_stat);
1524 if (!retval)
1525 retval = pid;
1526 } else {
1527 retval = wait_noreap_copyout(wo, p, pid, uid,
1528 CLD_CONTINUED, SIGCONT);
1529 BUG_ON(retval == 0);
1530 }
1531
1532 return retval;
1533}
1534
1535/*
1536 * Consider @p for a wait by @parent.
1537 *
1538 * -ECHILD should be in ->notask_error before the first call.
1539 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1540 * Returns zero if the search for a child should continue;
1541 * then ->notask_error is 0 if @p is an eligible child,
1542 * or another error from security_task_wait(), or still -ECHILD.
1543 */
1544static int wait_consider_task(struct wait_opts *wo, int ptrace,
1545 struct task_struct *p)
1546{
1547 int ret = eligible_child(wo, p);
1548 if (!ret)
1549 return ret;
1550
1551 ret = security_task_wait(p);
1552 if (unlikely(ret < 0)) {
1553 /*
1554 * If we have not yet seen any eligible child,
1555 * then let this error code replace -ECHILD.
1556 * A permission error will give the user a clue
1557 * to look for security policy problems, rather
1558 * than for mysterious wait bugs.
1559 */
1560 if (wo->notask_error)
1561 wo->notask_error = ret;
1562 return 0;
1563 }
1564
1565 /* dead body doesn't have much to contribute */
1566 if (unlikely(p->exit_state == EXIT_DEAD)) {
1567 /*
1568 * But do not ignore this task until the tracer does
1569 * wait_task_zombie()->do_notify_parent().
1570 */
1571 if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1572 wo->notask_error = 0;
1573 return 0;
1574 }
1575
1576 /* slay zombie? */
1577 if (p->exit_state == EXIT_ZOMBIE) {
1578 /*
1579 * A zombie ptracee is only visible to its ptracer.
1580 * Notification and reaping will be cascaded to the real
1581 * parent when the ptracer detaches.
1582 */
1583 if (likely(!ptrace) && unlikely(p->ptrace)) {
1584 /* it will become visible, clear notask_error */
1585 wo->notask_error = 0;
1586 return 0;
1587 }
1588
1589 /* we don't reap group leaders with subthreads */
1590 if (!delay_group_leader(p))
1591 return wait_task_zombie(wo, p);
1592
1593 /*
1594 * Allow access to stopped/continued state via zombie by
1595 * falling through. Clearing of notask_error is complex.
1596 *
1597 * When !@ptrace:
1598 *
1599 * If WEXITED is set, notask_error should naturally be
1600 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1601 * so, if there are live subthreads, there are events to
1602 * wait for. If all subthreads are dead, it's still safe
1603 * to clear - this function will be called again in finite
1604 * amount time once all the subthreads are released and
1605 * will then return without clearing.
1606 *
1607 * When @ptrace:
1608 *
1609 * Stopped state is per-task and thus can't change once the
1610 * target task dies. Only continued and exited can happen.
1611 * Clear notask_error if WCONTINUED | WEXITED.
1612 */
1613 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1614 wo->notask_error = 0;
1615 } else {
1616 /*
1617 * If @p is ptraced by a task in its real parent's group,
1618 * hide group stop/continued state when looking at @p as
1619 * the real parent; otherwise, a single stop can be
1620 * reported twice as group and ptrace stops.
1621 *
1622 * If a ptracer wants to distinguish the two events for its
1623 * own children, it should create a separate process which
1624 * takes the role of real parent.
1625 */
1626 if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1627 return 0;
1628
1629 /*
1630 * @p is alive and it's gonna stop, continue or exit, so
1631 * there always is something to wait for.
1632 */
1633 wo->notask_error = 0;
1634 }
1635
1636 /*
1637 * Wait for stopped. Depending on @ptrace, different stopped state
1638 * is used and the two don't interact with each other.
1639 */
1640 ret = wait_task_stopped(wo, ptrace, p);
1641 if (ret)
1642 return ret;
1643
1644 /*
1645 * Wait for continued. There's only one continued state and the
1646 * ptracer can consume it which can confuse the real parent. Don't
1647 * use WCONTINUED from ptracer. You don't need or want it.
1648 */
1649 return wait_task_continued(wo, p);
1650}
1651
1652/*
1653 * Do the work of do_wait() for one thread in the group, @tsk.
1654 *
1655 * -ECHILD should be in ->notask_error before the first call.
1656 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1657 * Returns zero if the search for a child should continue; then
1658 * ->notask_error is 0 if there were any eligible children,
1659 * or another error from security_task_wait(), or still -ECHILD.
1660 */
1661static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1662{
1663 struct task_struct *p;
1664
1665 list_for_each_entry(p, &tsk->children, sibling) {
1666 int ret = wait_consider_task(wo, 0, p);
1667 if (ret)
1668 return ret;
1669 }
1670
1671 return 0;
1672}
1673
1674static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1675{
1676 struct task_struct *p;
1677
1678 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1679 int ret = wait_consider_task(wo, 1, p);
1680 if (ret)
1681 return ret;
1682 }
1683
1684 return 0;
1685}
1686
1687static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1688 int sync, void *key)
1689{
1690 struct wait_opts *wo = container_of(wait, struct wait_opts,
1691 child_wait);
1692 struct task_struct *p = key;
1693
1694 if (!eligible_pid(wo, p))
1695 return 0;
1696
1697 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1698 return 0;
1699
1700 return default_wake_function(wait, mode, sync, key);
1701}
1702
1703void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1704{
1705 __wake_up_sync_key(&parent->signal->wait_chldexit,
1706 TASK_INTERRUPTIBLE, 1, p);
1707}
1708
1709static long do_wait(struct wait_opts *wo)
1710{
1711 struct task_struct *tsk;
1712 int retval;
1713
1714 trace_sched_process_wait(wo->wo_pid);
1715
1716 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1717 wo->child_wait.private = current;
1718 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1719repeat:
1720 /*
1721 * If there is nothing that can match our critiera just get out.
1722 * We will clear ->notask_error to zero if we see any child that
1723 * might later match our criteria, even if we are not able to reap
1724 * it yet.
1725 */
1726 wo->notask_error = -ECHILD;
1727 if ((wo->wo_type < PIDTYPE_MAX) &&
1728 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1729 goto notask;
1730
1731 set_current_state(TASK_INTERRUPTIBLE);
1732 read_lock(&tasklist_lock);
1733 tsk = current;
1734 do {
1735 retval = do_wait_thread(wo, tsk);
1736 if (retval)
1737 goto end;
1738
1739 retval = ptrace_do_wait(wo, tsk);
1740 if (retval)
1741 goto end;
1742
1743 if (wo->wo_flags & __WNOTHREAD)
1744 break;
1745 } while_each_thread(current, tsk);
1746 read_unlock(&tasklist_lock);
1747
1748notask:
1749 retval = wo->notask_error;
1750 if (!retval && !(wo->wo_flags & WNOHANG)) {
1751 retval = -ERESTARTSYS;
1752 if (!signal_pending(current)) {
1753 schedule();
1754 goto repeat;
1755 }
1756 }
1757end:
1758 __set_current_state(TASK_RUNNING);
1759 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1760 return retval;
1761}
1762
1763SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1764 infop, int, options, struct rusage __user *, ru)
1765{
1766 struct wait_opts wo;
1767 struct pid *pid = NULL;
1768 enum pid_type type;
1769 long ret;
1770
1771 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1772 return -EINVAL;
1773 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1774 return -EINVAL;
1775
1776 switch (which) {
1777 case P_ALL:
1778 type = PIDTYPE_MAX;
1779 break;
1780 case P_PID:
1781 type = PIDTYPE_PID;
1782 if (upid <= 0)
1783 return -EINVAL;
1784 break;
1785 case P_PGID:
1786 type = PIDTYPE_PGID;
1787 if (upid <= 0)
1788 return -EINVAL;
1789 break;
1790 default:
1791 return -EINVAL;
1792 }
1793
1794 if (type < PIDTYPE_MAX)
1795 pid = find_get_pid(upid);
1796
1797 wo.wo_type = type;
1798 wo.wo_pid = pid;
1799 wo.wo_flags = options;
1800 wo.wo_info = infop;
1801 wo.wo_stat = NULL;
1802 wo.wo_rusage = ru;
1803 ret = do_wait(&wo);
1804
1805 if (ret > 0) {
1806 ret = 0;
1807 } else if (infop) {
1808 /*
1809 * For a WNOHANG return, clear out all the fields
1810 * we would set so the user can easily tell the
1811 * difference.
1812 */
1813 if (!ret)
1814 ret = put_user(0, &infop->si_signo);
1815 if (!ret)
1816 ret = put_user(0, &infop->si_errno);
1817 if (!ret)
1818 ret = put_user(0, &infop->si_code);
1819 if (!ret)
1820 ret = put_user(0, &infop->si_pid);
1821 if (!ret)
1822 ret = put_user(0, &infop->si_uid);
1823 if (!ret)
1824 ret = put_user(0, &infop->si_status);
1825 }
1826
1827 put_pid(pid);
1828
1829 /* avoid REGPARM breakage on x86: */
1830 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1831 return ret;
1832}
1833
1834SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1835 int, options, struct rusage __user *, ru)
1836{
1837 struct wait_opts wo;
1838 struct pid *pid = NULL;
1839 enum pid_type type;
1840 long ret;
1841
1842 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1843 __WNOTHREAD|__WCLONE|__WALL))
1844 return -EINVAL;
1845
1846 if (upid == -1)
1847 type = PIDTYPE_MAX;
1848 else if (upid < 0) {
1849 type = PIDTYPE_PGID;
1850 pid = find_get_pid(-upid);
1851 } else if (upid == 0) {
1852 type = PIDTYPE_PGID;
1853 pid = get_task_pid(current, PIDTYPE_PGID);
1854 } else /* upid > 0 */ {
1855 type = PIDTYPE_PID;
1856 pid = find_get_pid(upid);
1857 }
1858
1859 wo.wo_type = type;
1860 wo.wo_pid = pid;
1861 wo.wo_flags = options | WEXITED;
1862 wo.wo_info = NULL;
1863 wo.wo_stat = stat_addr;
1864 wo.wo_rusage = ru;
1865 ret = do_wait(&wo);
1866 put_pid(pid);
1867
1868 /* avoid REGPARM breakage on x86: */
1869 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1870 return ret;
1871}
1872
1873#ifdef __ARCH_WANT_SYS_WAITPID
1874
1875/*
1876 * sys_waitpid() remains for compatibility. waitpid() should be
1877 * implemented by calling sys_wait4() from libc.a.
1878 */
1879SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1880{
1881 return sys_wait4(pid, stat_addr, options, NULL);
1882}
1883
1884#endif