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