4 * Copyright (C) 1991, 1992 Linus Torvalds
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>
58 #include <asm/uaccess.h>
59 #include <asm/unistd.h>
60 #include <asm/pgtable.h>
61 #include <asm/mmu_context.h>
63 static void __unhash_process(struct task_struct *p, bool group_dead)
66 detach_pid(p, PIDTYPE_PID);
68 detach_pid(p, PIDTYPE_PGID);
69 detach_pid(p, PIDTYPE_SID);
71 list_del_rcu(&p->tasks);
72 list_del_init(&p->sibling);
73 __this_cpu_dec(process_counts);
75 list_del_rcu(&p->thread_group);
76 list_del_rcu(&p->thread_node);
80 * This function expects the tasklist_lock write-locked.
82 static void __exit_signal(struct task_struct *tsk)
84 struct signal_struct *sig = tsk->signal;
85 bool group_dead = thread_group_leader(tsk);
86 struct sighand_struct *sighand;
87 struct tty_struct *uninitialized_var(tty);
88 cputime_t utime, stime;
90 sighand = rcu_dereference_check(tsk->sighand,
91 lockdep_tasklist_lock_is_held());
92 spin_lock(&sighand->siglock);
94 posix_cpu_timers_exit(tsk);
96 posix_cpu_timers_exit_group(tsk);
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.
105 if (unlikely(has_group_leader_pid(tsk)))
106 posix_cpu_timers_exit_group(tsk);
109 * If there is any task waiting for the group exit
112 if (sig->notify_count > 0 && !--sig->notify_count)
113 wake_up_process(sig->group_exit_task);
115 if (tsk == sig->curr_target)
116 sig->curr_target = next_thread(tsk);
120 * Accumulate here the counters for all threads as they die. We could
121 * skip the group leader because it is the last user of signal_struct,
122 * but we want to avoid the race with thread_group_cputime() which can
123 * see the empty ->thread_head list.
125 task_cputime(tsk, &utime, &stime);
126 write_seqlock(&sig->stats_lock);
129 sig->gtime += task_gtime(tsk);
130 sig->min_flt += tsk->min_flt;
131 sig->maj_flt += tsk->maj_flt;
132 sig->nvcsw += tsk->nvcsw;
133 sig->nivcsw += tsk->nivcsw;
134 sig->inblock += task_io_get_inblock(tsk);
135 sig->oublock += task_io_get_oublock(tsk);
136 task_io_accounting_add(&sig->ioac, &tsk->ioac);
137 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
139 __unhash_process(tsk, group_dead);
140 write_sequnlock(&sig->stats_lock);
143 * Do this under ->siglock, we can race with another thread
144 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
146 flush_sigqueue(&tsk->pending);
148 spin_unlock(&sighand->siglock);
150 __cleanup_sighand(sighand);
151 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
153 flush_sigqueue(&sig->shared_pending);
158 static void delayed_put_task_struct(struct rcu_head *rhp)
160 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
162 perf_event_delayed_put(tsk);
163 trace_sched_process_free(tsk);
164 put_task_struct(tsk);
168 void release_task(struct task_struct *p)
170 struct task_struct *leader;
173 /* don't need to get the RCU readlock here - the process is dead and
174 * can't be modifying its own credentials. But shut RCU-lockdep up */
176 atomic_dec(&__task_cred(p)->user->processes);
181 write_lock_irq(&tasklist_lock);
182 ptrace_release_task(p);
186 * If we are the last non-leader member of the thread
187 * group, and the leader is zombie, then notify the
188 * group leader's parent process. (if it wants notification.)
191 leader = p->group_leader;
192 if (leader != p && thread_group_empty(leader)
193 && leader->exit_state == EXIT_ZOMBIE) {
195 * If we were the last child thread and the leader has
196 * exited already, and the leader's parent ignores SIGCHLD,
197 * then we are the one who should release the leader.
199 zap_leader = do_notify_parent(leader, leader->exit_signal);
201 leader->exit_state = EXIT_DEAD;
204 write_unlock_irq(&tasklist_lock);
206 call_rcu(&p->rcu, delayed_put_task_struct);
209 if (unlikely(zap_leader))
214 * Determine if a process group is "orphaned", according to the POSIX
215 * definition in 2.2.2.52. Orphaned process groups are not to be affected
216 * by terminal-generated stop signals. Newly orphaned process groups are
217 * to receive a SIGHUP and a SIGCONT.
219 * "I ask you, have you ever known what it is to be an orphan?"
221 static int will_become_orphaned_pgrp(struct pid *pgrp,
222 struct task_struct *ignored_task)
224 struct task_struct *p;
226 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
227 if ((p == ignored_task) ||
228 (p->exit_state && thread_group_empty(p)) ||
229 is_global_init(p->real_parent))
232 if (task_pgrp(p->real_parent) != pgrp &&
233 task_session(p->real_parent) == task_session(p))
235 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
240 int is_current_pgrp_orphaned(void)
244 read_lock(&tasklist_lock);
245 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
246 read_unlock(&tasklist_lock);
251 static bool has_stopped_jobs(struct pid *pgrp)
253 struct task_struct *p;
255 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
256 if (p->signal->flags & SIGNAL_STOP_STOPPED)
258 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
264 * Check to see if any process groups have become orphaned as
265 * a result of our exiting, and if they have any stopped jobs,
266 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
269 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
271 struct pid *pgrp = task_pgrp(tsk);
272 struct task_struct *ignored_task = tsk;
275 /* exit: our father is in a different pgrp than
276 * we are and we were the only connection outside.
278 parent = tsk->real_parent;
280 /* reparent: our child is in a different pgrp than
281 * we are, and it was the only connection outside.
285 if (task_pgrp(parent) != pgrp &&
286 task_session(parent) == task_session(tsk) &&
287 will_become_orphaned_pgrp(pgrp, ignored_task) &&
288 has_stopped_jobs(pgrp)) {
289 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
290 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
296 * A task is exiting. If it owned this mm, find a new owner for the mm.
298 void mm_update_next_owner(struct mm_struct *mm)
300 struct task_struct *c, *g, *p = current;
304 * If the exiting or execing task is not the owner, it's
305 * someone else's problem.
310 * The current owner is exiting/execing and there are no other
311 * candidates. Do not leave the mm pointing to a possibly
312 * freed task structure.
314 if (atomic_read(&mm->mm_users) <= 1) {
319 read_lock(&tasklist_lock);
321 * Search in the children
323 list_for_each_entry(c, &p->children, sibling) {
325 goto assign_new_owner;
329 * Search in the siblings
331 list_for_each_entry(c, &p->real_parent->children, sibling) {
333 goto assign_new_owner;
337 * Search through everything else, we should not get here often.
339 for_each_process(g) {
340 if (g->flags & PF_KTHREAD)
342 for_each_thread(g, c) {
344 goto assign_new_owner;
349 read_unlock(&tasklist_lock);
351 * We found no owner yet mm_users > 1: this implies that we are
352 * most likely racing with swapoff (try_to_unuse()) or /proc or
353 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
362 * The task_lock protects c->mm from changing.
363 * We always want mm->owner->mm == mm
367 * Delay read_unlock() till we have the task_lock()
368 * to ensure that c does not slip away underneath us
370 read_unlock(&tasklist_lock);
380 #endif /* CONFIG_MEMCG */
383 * Turn us into a lazy TLB process if we
386 static void exit_mm(struct task_struct *tsk)
388 struct mm_struct *mm = tsk->mm;
389 struct core_state *core_state;
396 * Serialize with any possible pending coredump.
397 * We must hold mmap_sem around checking core_state
398 * and clearing tsk->mm. The core-inducing thread
399 * will increment ->nr_threads for each thread in the
400 * group with ->mm != NULL.
402 down_read(&mm->mmap_sem);
403 core_state = mm->core_state;
405 struct core_thread self;
407 up_read(&mm->mmap_sem);
410 self.next = xchg(&core_state->dumper.next, &self);
412 * Implies mb(), the result of xchg() must be visible
413 * to core_state->dumper.
415 if (atomic_dec_and_test(&core_state->nr_threads))
416 complete(&core_state->startup);
419 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
420 if (!self.task) /* see coredump_finish() */
422 freezable_schedule();
424 __set_task_state(tsk, TASK_RUNNING);
425 down_read(&mm->mmap_sem);
427 atomic_inc(&mm->mm_count);
428 BUG_ON(mm != tsk->active_mm);
429 /* more a memory barrier than a real lock */
432 up_read(&mm->mmap_sem);
433 enter_lazy_tlb(mm, current);
435 mm_update_next_owner(mm);
437 if (test_thread_flag(TIF_MEMDIE))
438 exit_oom_victim(tsk);
441 static struct task_struct *find_alive_thread(struct task_struct *p)
443 struct task_struct *t;
445 for_each_thread(p, t) {
446 if (!(t->flags & PF_EXITING))
452 static struct task_struct *find_child_reaper(struct task_struct *father)
453 __releases(&tasklist_lock)
454 __acquires(&tasklist_lock)
456 struct pid_namespace *pid_ns = task_active_pid_ns(father);
457 struct task_struct *reaper = pid_ns->child_reaper;
459 if (likely(reaper != father))
462 reaper = find_alive_thread(father);
464 pid_ns->child_reaper = reaper;
468 write_unlock_irq(&tasklist_lock);
469 if (unlikely(pid_ns == &init_pid_ns)) {
470 panic("Attempted to kill init! exitcode=0x%08x\n",
471 father->signal->group_exit_code ?: father->exit_code);
473 zap_pid_ns_processes(pid_ns);
474 write_lock_irq(&tasklist_lock);
480 * When we die, we re-parent all our children, and try to:
481 * 1. give them to another thread in our thread group, if such a member exists
482 * 2. give it to the first ancestor process which prctl'd itself as a
483 * child_subreaper for its children (like a service manager)
484 * 3. give it to the init process (PID 1) in our pid namespace
486 static struct task_struct *find_new_reaper(struct task_struct *father,
487 struct task_struct *child_reaper)
489 struct task_struct *thread, *reaper;
491 thread = find_alive_thread(father);
495 if (father->signal->has_child_subreaper) {
497 * Find the first ->is_child_subreaper ancestor in our pid_ns.
498 * We start from father to ensure we can not look into another
499 * namespace, this is safe because all its threads are dead.
501 for (reaper = father;
502 !same_thread_group(reaper, child_reaper);
503 reaper = reaper->real_parent) {
504 /* call_usermodehelper() descendants need this check */
505 if (reaper == &init_task)
507 if (!reaper->signal->is_child_subreaper)
509 thread = find_alive_thread(reaper);
519 * Any that need to be release_task'd are put on the @dead list.
521 static void reparent_leader(struct task_struct *father, struct task_struct *p,
522 struct list_head *dead)
524 if (unlikely(p->exit_state == EXIT_DEAD))
527 /* We don't want people slaying init. */
528 p->exit_signal = SIGCHLD;
530 /* If it has exited notify the new parent about this child's death. */
532 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
533 if (do_notify_parent(p, p->exit_signal)) {
534 p->exit_state = EXIT_DEAD;
535 list_add(&p->ptrace_entry, dead);
539 kill_orphaned_pgrp(p, father);
543 * This does two things:
545 * A. Make init inherit all the child processes
546 * B. Check to see if any process groups have become orphaned
547 * as a result of our exiting, and if they have any stopped
548 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
550 static void forget_original_parent(struct task_struct *father,
551 struct list_head *dead)
553 struct task_struct *p, *t, *reaper;
555 if (unlikely(!list_empty(&father->ptraced)))
556 exit_ptrace(father, dead);
558 /* Can drop and reacquire tasklist_lock */
559 reaper = find_child_reaper(father);
560 if (list_empty(&father->children))
563 reaper = find_new_reaper(father, reaper);
564 list_for_each_entry(p, &father->children, sibling) {
565 for_each_thread(p, t) {
566 t->real_parent = reaper;
567 BUG_ON((!t->ptrace) != (t->parent == father));
568 if (likely(!t->ptrace))
569 t->parent = t->real_parent;
570 if (t->pdeath_signal)
571 group_send_sig_info(t->pdeath_signal,
575 * If this is a threaded reparent there is no need to
576 * notify anyone anything has happened.
578 if (!same_thread_group(reaper, father))
579 reparent_leader(father, p, dead);
581 list_splice_tail_init(&father->children, &reaper->children);
585 * Send signals to all our closest relatives so that they know
586 * to properly mourn us..
588 static void exit_notify(struct task_struct *tsk, int group_dead)
591 struct task_struct *p, *n;
594 write_lock_irq(&tasklist_lock);
595 forget_original_parent(tsk, &dead);
598 kill_orphaned_pgrp(tsk->group_leader, NULL);
600 if (unlikely(tsk->ptrace)) {
601 int sig = thread_group_leader(tsk) &&
602 thread_group_empty(tsk) &&
603 !ptrace_reparented(tsk) ?
604 tsk->exit_signal : SIGCHLD;
605 autoreap = do_notify_parent(tsk, sig);
606 } else if (thread_group_leader(tsk)) {
607 autoreap = thread_group_empty(tsk) &&
608 do_notify_parent(tsk, tsk->exit_signal);
613 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
614 if (tsk->exit_state == EXIT_DEAD)
615 list_add(&tsk->ptrace_entry, &dead);
617 /* mt-exec, de_thread() is waiting for group leader */
618 if (unlikely(tsk->signal->notify_count < 0))
619 wake_up_process(tsk->signal->group_exit_task);
620 write_unlock_irq(&tasklist_lock);
622 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
623 list_del_init(&p->ptrace_entry);
628 #ifdef CONFIG_DEBUG_STACK_USAGE
629 static void check_stack_usage(void)
631 static DEFINE_SPINLOCK(low_water_lock);
632 static int lowest_to_date = THREAD_SIZE;
635 free = stack_not_used(current);
637 if (free >= lowest_to_date)
640 spin_lock(&low_water_lock);
641 if (free < lowest_to_date) {
642 pr_warn("%s (%d) used greatest stack depth: %lu bytes left\n",
643 current->comm, task_pid_nr(current), free);
644 lowest_to_date = free;
646 spin_unlock(&low_water_lock);
649 static inline void check_stack_usage(void) {}
652 void do_exit(long code)
654 struct task_struct *tsk = current;
656 TASKS_RCU(int tasks_rcu_i);
658 profile_task_exit(tsk);
661 WARN_ON(blk_needs_flush_plug(tsk));
663 if (unlikely(in_interrupt()))
664 panic("Aiee, killing interrupt handler!");
665 if (unlikely(!tsk->pid))
666 panic("Attempted to kill the idle task!");
669 * If do_exit is called because this processes oopsed, it's possible
670 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
671 * continuing. Amongst other possible reasons, this is to prevent
672 * mm_release()->clear_child_tid() from writing to a user-controlled
677 ptrace_event(PTRACE_EVENT_EXIT, code);
679 validate_creds_for_do_exit(tsk);
682 * We're taking recursive faults here in do_exit. Safest is to just
683 * leave this task alone and wait for reboot.
685 if (unlikely(tsk->flags & PF_EXITING)) {
686 pr_alert("Fixing recursive fault but reboot is needed!\n");
688 * We can do this unlocked here. The futex code uses
689 * this flag just to verify whether the pi state
690 * cleanup has been done or not. In the worst case it
691 * loops once more. We pretend that the cleanup was
692 * done as there is no way to return. Either the
693 * OWNER_DIED bit is set by now or we push the blocked
694 * task into the wait for ever nirwana as well.
696 tsk->flags |= PF_EXITPIDONE;
697 set_current_state(TASK_UNINTERRUPTIBLE);
701 exit_signals(tsk); /* sets PF_EXITING */
703 * Ensure that all new tsk->pi_lock acquisitions must observe
704 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
708 * Ensure that we must observe the pi_state in exit_mm() ->
709 * mm_release() -> exit_pi_state_list().
711 raw_spin_unlock_wait(&tsk->pi_lock);
713 if (unlikely(in_atomic())) {
714 pr_info("note: %s[%d] exited with preempt_count %d\n",
715 current->comm, task_pid_nr(current),
717 preempt_count_set(PREEMPT_ENABLED);
720 /* sync mm's RSS info before statistics gathering */
722 sync_mm_rss(tsk->mm);
723 acct_update_integrals(tsk);
724 group_dead = atomic_dec_and_test(&tsk->signal->live);
726 hrtimer_cancel(&tsk->signal->real_timer);
727 exit_itimers(tsk->signal);
729 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
731 acct_collect(code, group_dead);
736 tsk->exit_code = code;
737 taskstats_exit(tsk, group_dead);
743 trace_sched_process_exit(tsk);
750 disassociate_ctty(1);
751 exit_task_namespaces(tsk);
756 * Flush inherited counters to the parent - before the parent
757 * gets woken up by child-exit notifications.
759 * because of cgroup mode, must be called before cgroup_exit()
761 perf_event_exit_task(tsk);
766 * FIXME: do that only when needed, using sched_exit tracepoint
768 flush_ptrace_hw_breakpoint(tsk);
770 TASKS_RCU(preempt_disable());
771 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
772 TASKS_RCU(preempt_enable());
773 exit_notify(tsk, group_dead);
774 proc_exit_connector(tsk);
777 mpol_put(tsk->mempolicy);
778 tsk->mempolicy = NULL;
782 if (unlikely(current->pi_state_cache))
783 kfree(current->pi_state_cache);
786 * Make sure we are holding no locks:
788 debug_check_no_locks_held();
790 * We can do this unlocked here. The futex code uses this flag
791 * just to verify whether the pi state cleanup has been done
792 * or not. In the worst case it loops once more.
794 tsk->flags |= PF_EXITPIDONE;
797 exit_io_context(tsk);
799 if (tsk->splice_pipe)
800 free_pipe_info(tsk->splice_pipe);
802 if (tsk->task_frag.page)
803 put_page(tsk->task_frag.page);
805 validate_creds_for_do_exit(tsk);
810 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
812 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
815 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
816 * when the following two conditions become true.
817 * - There is race condition of mmap_sem (It is acquired by
819 * - SMI occurs before setting TASK_RUNINNG.
820 * (or hypervisor of virtual machine switches to other guest)
821 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
823 * To avoid it, we have to wait for releasing tsk->pi_lock which
824 * is held by try_to_wake_up()
827 raw_spin_unlock_wait(&tsk->pi_lock);
829 /* causes final put_task_struct in finish_task_switch(). */
830 tsk->state = TASK_DEAD;
831 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
834 /* Avoid "noreturn function does return". */
836 cpu_relax(); /* For when BUG is null */
838 EXPORT_SYMBOL_GPL(do_exit);
840 void complete_and_exit(struct completion *comp, long code)
847 EXPORT_SYMBOL(complete_and_exit);
849 SYSCALL_DEFINE1(exit, int, error_code)
851 do_exit((error_code&0xff)<<8);
855 * Take down every thread in the group. This is called by fatal signals
856 * as well as by sys_exit_group (below).
859 do_group_exit(int exit_code)
861 struct signal_struct *sig = current->signal;
863 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
865 if (signal_group_exit(sig))
866 exit_code = sig->group_exit_code;
867 else if (!thread_group_empty(current)) {
868 struct sighand_struct *const sighand = current->sighand;
870 spin_lock_irq(&sighand->siglock);
871 if (signal_group_exit(sig))
872 /* Another thread got here before we took the lock. */
873 exit_code = sig->group_exit_code;
875 sig->group_exit_code = exit_code;
876 sig->flags = SIGNAL_GROUP_EXIT;
877 zap_other_threads(current);
879 spin_unlock_irq(&sighand->siglock);
887 * this kills every thread in the thread group. Note that any externally
888 * wait4()-ing process will get the correct exit code - even if this
889 * thread is not the thread group leader.
891 SYSCALL_DEFINE1(exit_group, int, error_code)
893 do_group_exit((error_code & 0xff) << 8);
899 enum pid_type wo_type;
903 struct siginfo __user *wo_info;
905 struct rusage __user *wo_rusage;
907 wait_queue_t child_wait;
912 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
914 if (type != PIDTYPE_PID)
915 task = task->group_leader;
916 return task->pids[type].pid;
919 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
921 return wo->wo_type == PIDTYPE_MAX ||
922 task_pid_type(p, wo->wo_type) == wo->wo_pid;
926 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
928 if (!eligible_pid(wo, p))
932 * Wait for all children (clone and not) if __WALL is set or
933 * if it is traced by us.
935 if (ptrace || (wo->wo_flags & __WALL))
939 * Otherwise, wait for clone children *only* if __WCLONE is set;
940 * otherwise, wait for non-clone children *only*.
942 * Note: a "clone" child here is one that reports to its parent
943 * using a signal other than SIGCHLD, or a non-leader thread which
944 * we can only see if it is traced by us.
946 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
952 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
953 pid_t pid, uid_t uid, int why, int status)
955 struct siginfo __user *infop;
956 int retval = wo->wo_rusage
957 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
963 retval = put_user(SIGCHLD, &infop->si_signo);
965 retval = put_user(0, &infop->si_errno);
967 retval = put_user((short)why, &infop->si_code);
969 retval = put_user(pid, &infop->si_pid);
971 retval = put_user(uid, &infop->si_uid);
973 retval = put_user(status, &infop->si_status);
981 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
982 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
983 * the lock and this task is uninteresting. If we return nonzero, we have
984 * released the lock and the system call should return.
986 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
988 int state, retval, status;
989 pid_t pid = task_pid_vnr(p);
990 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
991 struct siginfo __user *infop;
993 if (!likely(wo->wo_flags & WEXITED))
996 if (unlikely(wo->wo_flags & WNOWAIT)) {
997 int exit_code = p->exit_code;
1001 read_unlock(&tasklist_lock);
1002 sched_annotate_sleep();
1004 if ((exit_code & 0x7f) == 0) {
1006 status = exit_code >> 8;
1008 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1009 status = exit_code & 0x7f;
1011 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1014 * Move the task's state to DEAD/TRACE, only one thread can do this.
1016 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1017 EXIT_TRACE : EXIT_DEAD;
1018 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1021 * We own this thread, nobody else can reap it.
1023 read_unlock(&tasklist_lock);
1024 sched_annotate_sleep();
1027 * Check thread_group_leader() to exclude the traced sub-threads.
1029 if (state == EXIT_DEAD && thread_group_leader(p)) {
1030 struct signal_struct *sig = p->signal;
1031 struct signal_struct *psig = current->signal;
1032 unsigned long maxrss;
1033 cputime_t tgutime, tgstime;
1036 * The resource counters for the group leader are in its
1037 * own task_struct. Those for dead threads in the group
1038 * are in its signal_struct, as are those for the child
1039 * processes it has previously reaped. All these
1040 * accumulate in the parent's signal_struct c* fields.
1042 * We don't bother to take a lock here to protect these
1043 * p->signal fields because the whole thread group is dead
1044 * and nobody can change them.
1046 * psig->stats_lock also protects us from our sub-theads
1047 * which can reap other children at the same time. Until
1048 * we change k_getrusage()-like users to rely on this lock
1049 * we have to take ->siglock as well.
1051 * We use thread_group_cputime_adjusted() to get times for
1052 * the thread group, which consolidates times for all threads
1053 * in the group including the group leader.
1055 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1056 spin_lock_irq(¤t->sighand->siglock);
1057 write_seqlock(&psig->stats_lock);
1058 psig->cutime += tgutime + sig->cutime;
1059 psig->cstime += tgstime + sig->cstime;
1060 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1062 p->min_flt + sig->min_flt + sig->cmin_flt;
1064 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1066 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1068 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1070 task_io_get_inblock(p) +
1071 sig->inblock + sig->cinblock;
1073 task_io_get_oublock(p) +
1074 sig->oublock + sig->coublock;
1075 maxrss = max(sig->maxrss, sig->cmaxrss);
1076 if (psig->cmaxrss < maxrss)
1077 psig->cmaxrss = maxrss;
1078 task_io_accounting_add(&psig->ioac, &p->ioac);
1079 task_io_accounting_add(&psig->ioac, &sig->ioac);
1080 write_sequnlock(&psig->stats_lock);
1081 spin_unlock_irq(¤t->sighand->siglock);
1084 retval = wo->wo_rusage
1085 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1086 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1087 ? p->signal->group_exit_code : p->exit_code;
1088 if (!retval && wo->wo_stat)
1089 retval = put_user(status, wo->wo_stat);
1091 infop = wo->wo_info;
1092 if (!retval && infop)
1093 retval = put_user(SIGCHLD, &infop->si_signo);
1094 if (!retval && infop)
1095 retval = put_user(0, &infop->si_errno);
1096 if (!retval && infop) {
1099 if ((status & 0x7f) == 0) {
1103 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1106 retval = put_user((short)why, &infop->si_code);
1108 retval = put_user(status, &infop->si_status);
1110 if (!retval && infop)
1111 retval = put_user(pid, &infop->si_pid);
1112 if (!retval && infop)
1113 retval = put_user(uid, &infop->si_uid);
1117 if (state == EXIT_TRACE) {
1118 write_lock_irq(&tasklist_lock);
1119 /* We dropped tasklist, ptracer could die and untrace */
1122 /* If parent wants a zombie, don't release it now */
1123 state = EXIT_ZOMBIE;
1124 if (do_notify_parent(p, p->exit_signal))
1126 p->exit_state = state;
1127 write_unlock_irq(&tasklist_lock);
1129 if (state == EXIT_DEAD)
1135 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1138 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1139 return &p->exit_code;
1141 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1142 return &p->signal->group_exit_code;
1148 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1150 * @ptrace: is the wait for ptrace
1151 * @p: task to wait for
1153 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1156 * read_lock(&tasklist_lock), which is released if return value is
1157 * non-zero. Also, grabs and releases @p->sighand->siglock.
1160 * 0 if wait condition didn't exist and search for other wait conditions
1161 * should continue. Non-zero return, -errno on failure and @p's pid on
1162 * success, implies that tasklist_lock is released and wait condition
1163 * search should terminate.
1165 static int wait_task_stopped(struct wait_opts *wo,
1166 int ptrace, struct task_struct *p)
1168 struct siginfo __user *infop;
1169 int retval, exit_code, *p_code, why;
1170 uid_t uid = 0; /* unneeded, required by compiler */
1174 * Traditionally we see ptrace'd stopped tasks regardless of options.
1176 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1179 if (!task_stopped_code(p, ptrace))
1183 spin_lock_irq(&p->sighand->siglock);
1185 p_code = task_stopped_code(p, ptrace);
1186 if (unlikely(!p_code))
1189 exit_code = *p_code;
1193 if (!unlikely(wo->wo_flags & WNOWAIT))
1196 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1198 spin_unlock_irq(&p->sighand->siglock);
1203 * Now we are pretty sure this task is interesting.
1204 * Make sure it doesn't get reaped out from under us while we
1205 * give up the lock and then examine it below. We don't want to
1206 * keep holding onto the tasklist_lock while we call getrusage and
1207 * possibly take page faults for user memory.
1210 pid = task_pid_vnr(p);
1211 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1212 read_unlock(&tasklist_lock);
1213 sched_annotate_sleep();
1215 if (unlikely(wo->wo_flags & WNOWAIT))
1216 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1218 retval = wo->wo_rusage
1219 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1220 if (!retval && wo->wo_stat)
1221 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1223 infop = wo->wo_info;
1224 if (!retval && infop)
1225 retval = put_user(SIGCHLD, &infop->si_signo);
1226 if (!retval && infop)
1227 retval = put_user(0, &infop->si_errno);
1228 if (!retval && infop)
1229 retval = put_user((short)why, &infop->si_code);
1230 if (!retval && infop)
1231 retval = put_user(exit_code, &infop->si_status);
1232 if (!retval && infop)
1233 retval = put_user(pid, &infop->si_pid);
1234 if (!retval && infop)
1235 retval = put_user(uid, &infop->si_uid);
1245 * Handle do_wait work for one task in a live, non-stopped state.
1246 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1247 * the lock and this task is uninteresting. If we return nonzero, we have
1248 * released the lock and the system call should return.
1250 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1256 if (!unlikely(wo->wo_flags & WCONTINUED))
1259 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1262 spin_lock_irq(&p->sighand->siglock);
1263 /* Re-check with the lock held. */
1264 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1265 spin_unlock_irq(&p->sighand->siglock);
1268 if (!unlikely(wo->wo_flags & WNOWAIT))
1269 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1270 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1271 spin_unlock_irq(&p->sighand->siglock);
1273 pid = task_pid_vnr(p);
1275 read_unlock(&tasklist_lock);
1276 sched_annotate_sleep();
1279 retval = wo->wo_rusage
1280 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1282 if (!retval && wo->wo_stat)
1283 retval = put_user(0xffff, wo->wo_stat);
1287 retval = wait_noreap_copyout(wo, p, pid, uid,
1288 CLD_CONTINUED, SIGCONT);
1289 BUG_ON(retval == 0);
1296 * Consider @p for a wait by @parent.
1298 * -ECHILD should be in ->notask_error before the first call.
1299 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1300 * Returns zero if the search for a child should continue;
1301 * then ->notask_error is 0 if @p is an eligible child,
1302 * or another error from security_task_wait(), or still -ECHILD.
1304 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1305 struct task_struct *p)
1308 * We can race with wait_task_zombie() from another thread.
1309 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1310 * can't confuse the checks below.
1312 int exit_state = ACCESS_ONCE(p->exit_state);
1315 if (unlikely(exit_state == EXIT_DEAD))
1318 ret = eligible_child(wo, ptrace, p);
1322 ret = security_task_wait(p);
1323 if (unlikely(ret < 0)) {
1325 * If we have not yet seen any eligible child,
1326 * then let this error code replace -ECHILD.
1327 * A permission error will give the user a clue
1328 * to look for security policy problems, rather
1329 * than for mysterious wait bugs.
1331 if (wo->notask_error)
1332 wo->notask_error = ret;
1336 if (unlikely(exit_state == EXIT_TRACE)) {
1338 * ptrace == 0 means we are the natural parent. In this case
1339 * we should clear notask_error, debugger will notify us.
1341 if (likely(!ptrace))
1342 wo->notask_error = 0;
1346 if (likely(!ptrace) && unlikely(p->ptrace)) {
1348 * If it is traced by its real parent's group, just pretend
1349 * the caller is ptrace_do_wait() and reap this child if it
1352 * This also hides group stop state from real parent; otherwise
1353 * a single stop can be reported twice as group and ptrace stop.
1354 * If a ptracer wants to distinguish these two events for its
1355 * own children it should create a separate process which takes
1356 * the role of real parent.
1358 if (!ptrace_reparented(p))
1363 if (exit_state == EXIT_ZOMBIE) {
1364 /* we don't reap group leaders with subthreads */
1365 if (!delay_group_leader(p)) {
1367 * A zombie ptracee is only visible to its ptracer.
1368 * Notification and reaping will be cascaded to the
1369 * real parent when the ptracer detaches.
1371 if (unlikely(ptrace) || likely(!p->ptrace))
1372 return wait_task_zombie(wo, p);
1376 * Allow access to stopped/continued state via zombie by
1377 * falling through. Clearing of notask_error is complex.
1381 * If WEXITED is set, notask_error should naturally be
1382 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1383 * so, if there are live subthreads, there are events to
1384 * wait for. If all subthreads are dead, it's still safe
1385 * to clear - this function will be called again in finite
1386 * amount time once all the subthreads are released and
1387 * will then return without clearing.
1391 * Stopped state is per-task and thus can't change once the
1392 * target task dies. Only continued and exited can happen.
1393 * Clear notask_error if WCONTINUED | WEXITED.
1395 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1396 wo->notask_error = 0;
1399 * @p is alive and it's gonna stop, continue or exit, so
1400 * there always is something to wait for.
1402 wo->notask_error = 0;
1406 * Wait for stopped. Depending on @ptrace, different stopped state
1407 * is used and the two don't interact with each other.
1409 ret = wait_task_stopped(wo, ptrace, p);
1414 * Wait for continued. There's only one continued state and the
1415 * ptracer can consume it which can confuse the real parent. Don't
1416 * use WCONTINUED from ptracer. You don't need or want it.
1418 return wait_task_continued(wo, p);
1422 * Do the work of do_wait() for one thread in the group, @tsk.
1424 * -ECHILD should be in ->notask_error before the first call.
1425 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1426 * Returns zero if the search for a child should continue; then
1427 * ->notask_error is 0 if there were any eligible children,
1428 * or another error from security_task_wait(), or still -ECHILD.
1430 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1432 struct task_struct *p;
1434 list_for_each_entry(p, &tsk->children, sibling) {
1435 int ret = wait_consider_task(wo, 0, p);
1444 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1446 struct task_struct *p;
1448 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1449 int ret = wait_consider_task(wo, 1, p);
1458 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1459 int sync, void *key)
1461 struct wait_opts *wo = container_of(wait, struct wait_opts,
1463 struct task_struct *p = key;
1465 if (!eligible_pid(wo, p))
1468 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1471 return default_wake_function(wait, mode, sync, key);
1474 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1476 __wake_up_sync_key(&parent->signal->wait_chldexit,
1477 TASK_INTERRUPTIBLE, 1, p);
1480 static long do_wait(struct wait_opts *wo)
1482 struct task_struct *tsk;
1485 trace_sched_process_wait(wo->wo_pid);
1487 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1488 wo->child_wait.private = current;
1489 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1492 * If there is nothing that can match our criteria, just get out.
1493 * We will clear ->notask_error to zero if we see any child that
1494 * might later match our criteria, even if we are not able to reap
1497 wo->notask_error = -ECHILD;
1498 if ((wo->wo_type < PIDTYPE_MAX) &&
1499 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1502 set_current_state(TASK_INTERRUPTIBLE);
1503 read_lock(&tasklist_lock);
1506 retval = do_wait_thread(wo, tsk);
1510 retval = ptrace_do_wait(wo, tsk);
1514 if (wo->wo_flags & __WNOTHREAD)
1516 } while_each_thread(current, tsk);
1517 read_unlock(&tasklist_lock);
1520 retval = wo->notask_error;
1521 if (!retval && !(wo->wo_flags & WNOHANG)) {
1522 retval = -ERESTARTSYS;
1523 if (!signal_pending(current)) {
1529 __set_current_state(TASK_RUNNING);
1530 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1534 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1535 infop, int, options, struct rusage __user *, ru)
1537 struct wait_opts wo;
1538 struct pid *pid = NULL;
1542 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1543 __WNOTHREAD|__WCLONE|__WALL))
1545 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1558 type = PIDTYPE_PGID;
1566 if (type < PIDTYPE_MAX)
1567 pid = find_get_pid(upid);
1571 wo.wo_flags = options;
1581 * For a WNOHANG return, clear out all the fields
1582 * we would set so the user can easily tell the
1586 ret = put_user(0, &infop->si_signo);
1588 ret = put_user(0, &infop->si_errno);
1590 ret = put_user(0, &infop->si_code);
1592 ret = put_user(0, &infop->si_pid);
1594 ret = put_user(0, &infop->si_uid);
1596 ret = put_user(0, &infop->si_status);
1603 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1604 int, options, struct rusage __user *, ru)
1606 struct wait_opts wo;
1607 struct pid *pid = NULL;
1611 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1612 __WNOTHREAD|__WCLONE|__WALL))
1617 else if (upid < 0) {
1618 type = PIDTYPE_PGID;
1619 pid = find_get_pid(-upid);
1620 } else if (upid == 0) {
1621 type = PIDTYPE_PGID;
1622 pid = get_task_pid(current, PIDTYPE_PGID);
1623 } else /* upid > 0 */ {
1625 pid = find_get_pid(upid);
1630 wo.wo_flags = options | WEXITED;
1632 wo.wo_stat = stat_addr;
1640 #ifdef __ARCH_WANT_SYS_WAITPID
1643 * sys_waitpid() remains for compatibility. waitpid() should be
1644 * implemented by calling sys_wait4() from libc.a.
1646 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1648 return sys_wait4(pid, stat_addr, options, NULL);