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 * Note that if this function returns a valid task_struct pointer (!NULL)
215 * task->usage must remain >0 for the duration of the RCU critical section.
217 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
219 struct sighand_struct *sighand;
220 struct task_struct *task;
223 * We need to verify that release_task() was not called and thus
224 * delayed_put_task_struct() can't run and drop the last reference
225 * before rcu_read_unlock(). We check task->sighand != NULL,
226 * but we can read the already freed and reused memory.
229 task = rcu_dereference(*ptask);
233 probe_kernel_address(&task->sighand, sighand);
236 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
237 * was already freed we can not miss the preceding update of this
241 if (unlikely(task != READ_ONCE(*ptask)))
245 * We've re-checked that "task == *ptask", now we have two different
248 * 1. This is actually the same task/task_struct. In this case
249 * sighand != NULL tells us it is still alive.
251 * 2. This is another task which got the same memory for task_struct.
252 * We can't know this of course, and we can not trust
255 * In this case we actually return a random value, but this is
258 * If we return NULL - we can pretend that we actually noticed that
259 * *ptask was updated when the previous task has exited. Or pretend
260 * that probe_slab_address(&sighand) reads NULL.
262 * If we return the new task (because sighand is not NULL for any
263 * reason) - this is fine too. This (new) task can't go away before
266 * And note: We could even eliminate the false positive if re-read
267 * task->sighand once again to avoid the falsely NULL. But this case
268 * is very unlikely so we don't care.
276 struct task_struct *try_get_task_struct(struct task_struct **ptask)
278 struct task_struct *task;
281 task = task_rcu_dereference(ptask);
283 get_task_struct(task);
290 * Determine if a process group is "orphaned", according to the POSIX
291 * definition in 2.2.2.52. Orphaned process groups are not to be affected
292 * by terminal-generated stop signals. Newly orphaned process groups are
293 * to receive a SIGHUP and a SIGCONT.
295 * "I ask you, have you ever known what it is to be an orphan?"
297 static int will_become_orphaned_pgrp(struct pid *pgrp,
298 struct task_struct *ignored_task)
300 struct task_struct *p;
302 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
303 if ((p == ignored_task) ||
304 (p->exit_state && thread_group_empty(p)) ||
305 is_global_init(p->real_parent))
308 if (task_pgrp(p->real_parent) != pgrp &&
309 task_session(p->real_parent) == task_session(p))
311 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
316 int is_current_pgrp_orphaned(void)
320 read_lock(&tasklist_lock);
321 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
322 read_unlock(&tasklist_lock);
327 static bool has_stopped_jobs(struct pid *pgrp)
329 struct task_struct *p;
331 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
332 if (p->signal->flags & SIGNAL_STOP_STOPPED)
334 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
340 * Check to see if any process groups have become orphaned as
341 * a result of our exiting, and if they have any stopped jobs,
342 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
345 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
347 struct pid *pgrp = task_pgrp(tsk);
348 struct task_struct *ignored_task = tsk;
351 /* exit: our father is in a different pgrp than
352 * we are and we were the only connection outside.
354 parent = tsk->real_parent;
356 /* reparent: our child is in a different pgrp than
357 * we are, and it was the only connection outside.
361 if (task_pgrp(parent) != pgrp &&
362 task_session(parent) == task_session(tsk) &&
363 will_become_orphaned_pgrp(pgrp, ignored_task) &&
364 has_stopped_jobs(pgrp)) {
365 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
366 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
372 * A task is exiting. If it owned this mm, find a new owner for the mm.
374 void mm_update_next_owner(struct mm_struct *mm)
376 struct task_struct *c, *g, *p = current;
380 * If the exiting or execing task is not the owner, it's
381 * someone else's problem.
386 * The current owner is exiting/execing and there are no other
387 * candidates. Do not leave the mm pointing to a possibly
388 * freed task structure.
390 if (atomic_read(&mm->mm_users) <= 1) {
395 read_lock(&tasklist_lock);
397 * Search in the children
399 list_for_each_entry(c, &p->children, sibling) {
401 goto assign_new_owner;
405 * Search in the siblings
407 list_for_each_entry(c, &p->real_parent->children, sibling) {
409 goto assign_new_owner;
413 * Search through everything else, we should not get here often.
415 for_each_process(g) {
416 if (g->flags & PF_KTHREAD)
418 for_each_thread(g, c) {
420 goto assign_new_owner;
425 read_unlock(&tasklist_lock);
427 * We found no owner yet mm_users > 1: this implies that we are
428 * most likely racing with swapoff (try_to_unuse()) or /proc or
429 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
438 * The task_lock protects c->mm from changing.
439 * We always want mm->owner->mm == mm
443 * Delay read_unlock() till we have the task_lock()
444 * to ensure that c does not slip away underneath us
446 read_unlock(&tasklist_lock);
456 #endif /* CONFIG_MEMCG */
459 * Turn us into a lazy TLB process if we
462 static void exit_mm(struct task_struct *tsk)
464 struct mm_struct *mm = tsk->mm;
465 struct core_state *core_state;
472 * Serialize with any possible pending coredump.
473 * We must hold mmap_sem around checking core_state
474 * and clearing tsk->mm. The core-inducing thread
475 * will increment ->nr_threads for each thread in the
476 * group with ->mm != NULL.
478 down_read(&mm->mmap_sem);
479 core_state = mm->core_state;
481 struct core_thread self;
483 up_read(&mm->mmap_sem);
486 self.next = xchg(&core_state->dumper.next, &self);
488 * Implies mb(), the result of xchg() must be visible
489 * to core_state->dumper.
491 if (atomic_dec_and_test(&core_state->nr_threads))
492 complete(&core_state->startup);
495 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
496 if (!self.task) /* see coredump_finish() */
498 freezable_schedule();
500 __set_task_state(tsk, TASK_RUNNING);
501 down_read(&mm->mmap_sem);
503 atomic_inc(&mm->mm_count);
504 BUG_ON(mm != tsk->active_mm);
505 /* more a memory barrier than a real lock */
508 up_read(&mm->mmap_sem);
509 enter_lazy_tlb(mm, current);
511 mm_update_next_owner(mm);
513 if (test_thread_flag(TIF_MEMDIE))
517 static struct task_struct *find_alive_thread(struct task_struct *p)
519 struct task_struct *t;
521 for_each_thread(p, t) {
522 if (!(t->flags & PF_EXITING))
528 static struct task_struct *find_child_reaper(struct task_struct *father,
529 struct list_head *dead)
530 __releases(&tasklist_lock)
531 __acquires(&tasklist_lock)
533 struct pid_namespace *pid_ns = task_active_pid_ns(father);
534 struct task_struct *reaper = pid_ns->child_reaper;
535 struct task_struct *p, *n;
537 if (likely(reaper != father))
540 reaper = find_alive_thread(father);
542 pid_ns->child_reaper = reaper;
546 write_unlock_irq(&tasklist_lock);
547 if (unlikely(pid_ns == &init_pid_ns)) {
548 panic("Attempted to kill init! exitcode=0x%08x\n",
549 father->signal->group_exit_code ?: father->exit_code);
552 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
553 list_del_init(&p->ptrace_entry);
557 zap_pid_ns_processes(pid_ns);
558 write_lock_irq(&tasklist_lock);
564 * When we die, we re-parent all our children, and try to:
565 * 1. give them to another thread in our thread group, if such a member exists
566 * 2. give it to the first ancestor process which prctl'd itself as a
567 * child_subreaper for its children (like a service manager)
568 * 3. give it to the init process (PID 1) in our pid namespace
570 static struct task_struct *find_new_reaper(struct task_struct *father,
571 struct task_struct *child_reaper)
573 struct task_struct *thread, *reaper;
575 thread = find_alive_thread(father);
579 if (father->signal->has_child_subreaper) {
581 * Find the first ->is_child_subreaper ancestor in our pid_ns.
582 * We start from father to ensure we can not look into another
583 * namespace, this is safe because all its threads are dead.
585 for (reaper = father;
586 !same_thread_group(reaper, child_reaper);
587 reaper = reaper->real_parent) {
588 /* call_usermodehelper() descendants need this check */
589 if (reaper == &init_task)
591 if (!reaper->signal->is_child_subreaper)
593 thread = find_alive_thread(reaper);
603 * Any that need to be release_task'd are put on the @dead list.
605 static void reparent_leader(struct task_struct *father, struct task_struct *p,
606 struct list_head *dead)
608 if (unlikely(p->exit_state == EXIT_DEAD))
611 /* We don't want people slaying init. */
612 p->exit_signal = SIGCHLD;
614 /* If it has exited notify the new parent about this child's death. */
616 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
617 if (do_notify_parent(p, p->exit_signal)) {
618 p->exit_state = EXIT_DEAD;
619 list_add(&p->ptrace_entry, dead);
623 kill_orphaned_pgrp(p, father);
627 * This does two things:
629 * A. Make init inherit all the child processes
630 * B. Check to see if any process groups have become orphaned
631 * as a result of our exiting, and if they have any stopped
632 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
634 static void forget_original_parent(struct task_struct *father,
635 struct list_head *dead)
637 struct task_struct *p, *t, *reaper;
639 if (unlikely(!list_empty(&father->ptraced)))
640 exit_ptrace(father, dead);
642 /* Can drop and reacquire tasklist_lock */
643 reaper = find_child_reaper(father, dead);
644 if (list_empty(&father->children))
647 reaper = find_new_reaper(father, reaper);
648 list_for_each_entry(p, &father->children, sibling) {
649 for_each_thread(p, t) {
650 t->real_parent = reaper;
651 BUG_ON((!t->ptrace) != (t->parent == father));
652 if (likely(!t->ptrace))
653 t->parent = t->real_parent;
654 if (t->pdeath_signal)
655 group_send_sig_info(t->pdeath_signal,
659 * If this is a threaded reparent there is no need to
660 * notify anyone anything has happened.
662 if (!same_thread_group(reaper, father))
663 reparent_leader(father, p, dead);
665 list_splice_tail_init(&father->children, &reaper->children);
669 * Send signals to all our closest relatives so that they know
670 * to properly mourn us..
672 static void exit_notify(struct task_struct *tsk, int group_dead)
675 struct task_struct *p, *n;
678 write_lock_irq(&tasklist_lock);
679 forget_original_parent(tsk, &dead);
682 kill_orphaned_pgrp(tsk->group_leader, NULL);
684 if (unlikely(tsk->ptrace)) {
685 int sig = thread_group_leader(tsk) &&
686 thread_group_empty(tsk) &&
687 !ptrace_reparented(tsk) ?
688 tsk->exit_signal : SIGCHLD;
689 autoreap = do_notify_parent(tsk, sig);
690 } else if (thread_group_leader(tsk)) {
691 autoreap = thread_group_empty(tsk) &&
692 do_notify_parent(tsk, tsk->exit_signal);
697 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
698 if (tsk->exit_state == EXIT_DEAD)
699 list_add(&tsk->ptrace_entry, &dead);
701 /* mt-exec, de_thread() is waiting for group leader */
702 if (unlikely(tsk->signal->notify_count < 0))
703 wake_up_process(tsk->signal->group_exit_task);
704 write_unlock_irq(&tasklist_lock);
706 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
707 list_del_init(&p->ptrace_entry);
712 #ifdef CONFIG_DEBUG_STACK_USAGE
713 static void check_stack_usage(void)
715 static DEFINE_SPINLOCK(low_water_lock);
716 static int lowest_to_date = THREAD_SIZE;
719 free = stack_not_used(current);
721 if (free >= lowest_to_date)
724 spin_lock(&low_water_lock);
725 if (free < lowest_to_date) {
726 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
727 current->comm, task_pid_nr(current), free);
728 lowest_to_date = free;
730 spin_unlock(&low_water_lock);
733 static inline void check_stack_usage(void) {}
736 void __noreturn do_exit(long code)
738 struct task_struct *tsk = current;
740 TASKS_RCU(int tasks_rcu_i);
742 profile_task_exit(tsk);
745 WARN_ON(blk_needs_flush_plug(tsk));
747 if (unlikely(in_interrupt()))
748 panic("Aiee, killing interrupt handler!");
749 if (unlikely(!tsk->pid))
750 panic("Attempted to kill the idle task!");
753 * If do_exit is called because this processes oopsed, it's possible
754 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
755 * continuing. Amongst other possible reasons, this is to prevent
756 * mm_release()->clear_child_tid() from writing to a user-controlled
761 ptrace_event(PTRACE_EVENT_EXIT, code);
763 validate_creds_for_do_exit(tsk);
766 * We're taking recursive faults here in do_exit. Safest is to just
767 * leave this task alone and wait for reboot.
769 if (unlikely(tsk->flags & PF_EXITING)) {
770 pr_alert("Fixing recursive fault but reboot is needed!\n");
772 * We can do this unlocked here. The futex code uses
773 * this flag just to verify whether the pi state
774 * cleanup has been done or not. In the worst case it
775 * loops once more. We pretend that the cleanup was
776 * done as there is no way to return. Either the
777 * OWNER_DIED bit is set by now or we push the blocked
778 * task into the wait for ever nirwana as well.
780 tsk->flags |= PF_EXITPIDONE;
781 set_current_state(TASK_UNINTERRUPTIBLE);
785 exit_signals(tsk); /* sets PF_EXITING */
787 * Ensure that all new tsk->pi_lock acquisitions must observe
788 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
792 * Ensure that we must observe the pi_state in exit_mm() ->
793 * mm_release() -> exit_pi_state_list().
795 raw_spin_unlock_wait(&tsk->pi_lock);
797 if (unlikely(in_atomic())) {
798 pr_info("note: %s[%d] exited with preempt_count %d\n",
799 current->comm, task_pid_nr(current),
801 preempt_count_set(PREEMPT_ENABLED);
804 /* sync mm's RSS info before statistics gathering */
806 sync_mm_rss(tsk->mm);
807 acct_update_integrals(tsk);
808 group_dead = atomic_dec_and_test(&tsk->signal->live);
810 hrtimer_cancel(&tsk->signal->real_timer);
811 exit_itimers(tsk->signal);
813 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
815 acct_collect(code, group_dead);
820 tsk->exit_code = code;
821 taskstats_exit(tsk, group_dead);
827 trace_sched_process_exit(tsk);
834 disassociate_ctty(1);
835 exit_task_namespaces(tsk);
840 * Flush inherited counters to the parent - before the parent
841 * gets woken up by child-exit notifications.
843 * because of cgroup mode, must be called before cgroup_exit()
845 perf_event_exit_task(tsk);
847 sched_autogroup_exit_task(tsk);
851 * FIXME: do that only when needed, using sched_exit tracepoint
853 flush_ptrace_hw_breakpoint(tsk);
855 TASKS_RCU(preempt_disable());
856 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
857 TASKS_RCU(preempt_enable());
858 exit_notify(tsk, group_dead);
859 proc_exit_connector(tsk);
860 mpol_put_task_policy(tsk);
862 if (unlikely(current->pi_state_cache))
863 kfree(current->pi_state_cache);
866 * Make sure we are holding no locks:
868 debug_check_no_locks_held();
870 * We can do this unlocked here. The futex code uses this flag
871 * just to verify whether the pi state cleanup has been done
872 * or not. In the worst case it loops once more.
874 tsk->flags |= PF_EXITPIDONE;
877 exit_io_context(tsk);
879 if (tsk->splice_pipe)
880 free_pipe_info(tsk->splice_pipe);
882 if (tsk->task_frag.page)
883 put_page(tsk->task_frag.page);
885 validate_creds_for_do_exit(tsk);
890 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
892 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
896 EXPORT_SYMBOL_GPL(do_exit);
898 void complete_and_exit(struct completion *comp, long code)
905 EXPORT_SYMBOL(complete_and_exit);
907 SYSCALL_DEFINE1(exit, int, error_code)
909 do_exit((error_code&0xff)<<8);
913 * Take down every thread in the group. This is called by fatal signals
914 * as well as by sys_exit_group (below).
917 do_group_exit(int exit_code)
919 struct signal_struct *sig = current->signal;
921 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
923 if (signal_group_exit(sig))
924 exit_code = sig->group_exit_code;
925 else if (!thread_group_empty(current)) {
926 struct sighand_struct *const sighand = current->sighand;
928 spin_lock_irq(&sighand->siglock);
929 if (signal_group_exit(sig))
930 /* Another thread got here before we took the lock. */
931 exit_code = sig->group_exit_code;
933 sig->group_exit_code = exit_code;
934 sig->flags = SIGNAL_GROUP_EXIT;
935 zap_other_threads(current);
937 spin_unlock_irq(&sighand->siglock);
945 * this kills every thread in the thread group. Note that any externally
946 * wait4()-ing process will get the correct exit code - even if this
947 * thread is not the thread group leader.
949 SYSCALL_DEFINE1(exit_group, int, error_code)
951 do_group_exit((error_code & 0xff) << 8);
957 enum pid_type wo_type;
961 struct siginfo __user *wo_info;
963 struct rusage __user *wo_rusage;
965 wait_queue_t child_wait;
970 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
972 if (type != PIDTYPE_PID)
973 task = task->group_leader;
974 return task->pids[type].pid;
977 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
979 return wo->wo_type == PIDTYPE_MAX ||
980 task_pid_type(p, wo->wo_type) == wo->wo_pid;
984 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
986 if (!eligible_pid(wo, p))
990 * Wait for all children (clone and not) if __WALL is set or
991 * if it is traced by us.
993 if (ptrace || (wo->wo_flags & __WALL))
997 * Otherwise, wait for clone children *only* if __WCLONE is set;
998 * otherwise, wait for non-clone children *only*.
1000 * Note: a "clone" child here is one that reports to its parent
1001 * using a signal other than SIGCHLD, or a non-leader thread which
1002 * we can only see if it is traced by us.
1004 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1010 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1011 pid_t pid, uid_t uid, int why, int status)
1013 struct siginfo __user *infop;
1014 int retval = wo->wo_rusage
1015 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1018 infop = wo->wo_info;
1021 retval = put_user(SIGCHLD, &infop->si_signo);
1023 retval = put_user(0, &infop->si_errno);
1025 retval = put_user((short)why, &infop->si_code);
1027 retval = put_user(pid, &infop->si_pid);
1029 retval = put_user(uid, &infop->si_uid);
1031 retval = put_user(status, &infop->si_status);
1039 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1040 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1041 * the lock and this task is uninteresting. If we return nonzero, we have
1042 * released the lock and the system call should return.
1044 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1046 int state, retval, status;
1047 pid_t pid = task_pid_vnr(p);
1048 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1049 struct siginfo __user *infop;
1051 if (!likely(wo->wo_flags & WEXITED))
1054 if (unlikely(wo->wo_flags & WNOWAIT)) {
1055 int exit_code = p->exit_code;
1059 read_unlock(&tasklist_lock);
1060 sched_annotate_sleep();
1062 if ((exit_code & 0x7f) == 0) {
1064 status = exit_code >> 8;
1066 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1067 status = exit_code & 0x7f;
1069 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1072 * Move the task's state to DEAD/TRACE, only one thread can do this.
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)
1079 * We own this thread, nobody else can reap it.
1081 read_unlock(&tasklist_lock);
1082 sched_annotate_sleep();
1085 * Check thread_group_leader() to exclude the traced sub-threads.
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 cputime_t tgutime, tgstime;
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.
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.
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.
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.
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;
1120 p->min_flt + sig->min_flt + sig->cmin_flt;
1122 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1124 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1126 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1128 task_io_get_inblock(p) +
1129 sig->inblock + sig->cinblock;
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);
1142 retval = wo->wo_rusage
1143 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1144 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1145 ? p->signal->group_exit_code : p->exit_code;
1146 if (!retval && wo->wo_stat)
1147 retval = put_user(status, wo->wo_stat);
1149 infop = wo->wo_info;
1150 if (!retval && infop)
1151 retval = put_user(SIGCHLD, &infop->si_signo);
1152 if (!retval && infop)
1153 retval = put_user(0, &infop->si_errno);
1154 if (!retval && infop) {
1157 if ((status & 0x7f) == 0) {
1161 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1164 retval = put_user((short)why, &infop->si_code);
1166 retval = put_user(status, &infop->si_status);
1168 if (!retval && infop)
1169 retval = put_user(pid, &infop->si_pid);
1170 if (!retval && infop)
1171 retval = put_user(uid, &infop->si_uid);
1175 if (state == EXIT_TRACE) {
1176 write_lock_irq(&tasklist_lock);
1177 /* We dropped tasklist, ptracer could die and untrace */
1180 /* If parent wants a zombie, don't release it now */
1181 state = EXIT_ZOMBIE;
1182 if (do_notify_parent(p, p->exit_signal))
1184 p->exit_state = state;
1185 write_unlock_irq(&tasklist_lock);
1187 if (state == EXIT_DEAD)
1193 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1196 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1197 return &p->exit_code;
1199 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1200 return &p->signal->group_exit_code;
1206 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1208 * @ptrace: is the wait for ptrace
1209 * @p: task to wait for
1211 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1214 * read_lock(&tasklist_lock), which is released if return value is
1215 * non-zero. Also, grabs and releases @p->sighand->siglock.
1218 * 0 if wait condition didn't exist and search for other wait conditions
1219 * should continue. Non-zero return, -errno on failure and @p's pid on
1220 * success, implies that tasklist_lock is released and wait condition
1221 * search should terminate.
1223 static int wait_task_stopped(struct wait_opts *wo,
1224 int ptrace, struct task_struct *p)
1226 struct siginfo __user *infop;
1227 int retval, exit_code, *p_code, why;
1228 uid_t uid = 0; /* unneeded, required by compiler */
1232 * Traditionally we see ptrace'd stopped tasks regardless of options.
1234 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1237 if (!task_stopped_code(p, ptrace))
1241 spin_lock_irq(&p->sighand->siglock);
1243 p_code = task_stopped_code(p, ptrace);
1244 if (unlikely(!p_code))
1247 exit_code = *p_code;
1251 if (!unlikely(wo->wo_flags & WNOWAIT))
1254 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1256 spin_unlock_irq(&p->sighand->siglock);
1261 * Now we are pretty sure this task is interesting.
1262 * Make sure it doesn't get reaped out from under us while we
1263 * give up the lock and then examine it below. We don't want to
1264 * keep holding onto the tasklist_lock while we call getrusage and
1265 * possibly take page faults for user memory.
1268 pid = task_pid_vnr(p);
1269 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1270 read_unlock(&tasklist_lock);
1271 sched_annotate_sleep();
1273 if (unlikely(wo->wo_flags & WNOWAIT))
1274 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1276 retval = wo->wo_rusage
1277 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1278 if (!retval && wo->wo_stat)
1279 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1281 infop = wo->wo_info;
1282 if (!retval && infop)
1283 retval = put_user(SIGCHLD, &infop->si_signo);
1284 if (!retval && infop)
1285 retval = put_user(0, &infop->si_errno);
1286 if (!retval && infop)
1287 retval = put_user((short)why, &infop->si_code);
1288 if (!retval && infop)
1289 retval = put_user(exit_code, &infop->si_status);
1290 if (!retval && infop)
1291 retval = put_user(pid, &infop->si_pid);
1292 if (!retval && infop)
1293 retval = put_user(uid, &infop->si_uid);
1303 * Handle do_wait work for one task in a live, non-stopped state.
1304 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1305 * the lock and this task is uninteresting. If we return nonzero, we have
1306 * released the lock and the system call should return.
1308 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1314 if (!unlikely(wo->wo_flags & WCONTINUED))
1317 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1320 spin_lock_irq(&p->sighand->siglock);
1321 /* Re-check with the lock held. */
1322 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1323 spin_unlock_irq(&p->sighand->siglock);
1326 if (!unlikely(wo->wo_flags & WNOWAIT))
1327 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1328 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1329 spin_unlock_irq(&p->sighand->siglock);
1331 pid = task_pid_vnr(p);
1333 read_unlock(&tasklist_lock);
1334 sched_annotate_sleep();
1337 retval = wo->wo_rusage
1338 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1340 if (!retval && wo->wo_stat)
1341 retval = put_user(0xffff, wo->wo_stat);
1345 retval = wait_noreap_copyout(wo, p, pid, uid,
1346 CLD_CONTINUED, SIGCONT);
1347 BUG_ON(retval == 0);
1354 * Consider @p for a wait by @parent.
1356 * -ECHILD should be in ->notask_error before the first call.
1357 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1358 * Returns zero if the search for a child should continue;
1359 * then ->notask_error is 0 if @p is an eligible child,
1360 * or another error from security_task_wait(), or still -ECHILD.
1362 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1363 struct task_struct *p)
1366 * We can race with wait_task_zombie() from another thread.
1367 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1368 * can't confuse the checks below.
1370 int exit_state = ACCESS_ONCE(p->exit_state);
1373 if (unlikely(exit_state == EXIT_DEAD))
1376 ret = eligible_child(wo, ptrace, p);
1380 ret = security_task_wait(p);
1381 if (unlikely(ret < 0)) {
1383 * If we have not yet seen any eligible child,
1384 * then let this error code replace -ECHILD.
1385 * A permission error will give the user a clue
1386 * to look for security policy problems, rather
1387 * than for mysterious wait bugs.
1389 if (wo->notask_error)
1390 wo->notask_error = ret;
1394 if (unlikely(exit_state == EXIT_TRACE)) {
1396 * ptrace == 0 means we are the natural parent. In this case
1397 * we should clear notask_error, debugger will notify us.
1399 if (likely(!ptrace))
1400 wo->notask_error = 0;
1404 if (likely(!ptrace) && unlikely(p->ptrace)) {
1406 * If it is traced by its real parent's group, just pretend
1407 * the caller is ptrace_do_wait() and reap this child if it
1410 * This also hides group stop state from real parent; otherwise
1411 * a single stop can be reported twice as group and ptrace stop.
1412 * If a ptracer wants to distinguish these two events for its
1413 * own children it should create a separate process which takes
1414 * the role of real parent.
1416 if (!ptrace_reparented(p))
1421 if (exit_state == EXIT_ZOMBIE) {
1422 /* we don't reap group leaders with subthreads */
1423 if (!delay_group_leader(p)) {
1425 * A zombie ptracee is only visible to its ptracer.
1426 * Notification and reaping will be cascaded to the
1427 * real parent when the ptracer detaches.
1429 if (unlikely(ptrace) || likely(!p->ptrace))
1430 return wait_task_zombie(wo, p);
1434 * Allow access to stopped/continued state via zombie by
1435 * falling through. Clearing of notask_error is complex.
1439 * If WEXITED is set, notask_error should naturally be
1440 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1441 * so, if there are live subthreads, there are events to
1442 * wait for. If all subthreads are dead, it's still safe
1443 * to clear - this function will be called again in finite
1444 * amount time once all the subthreads are released and
1445 * will then return without clearing.
1449 * Stopped state is per-task and thus can't change once the
1450 * target task dies. Only continued and exited can happen.
1451 * Clear notask_error if WCONTINUED | WEXITED.
1453 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1454 wo->notask_error = 0;
1457 * @p is alive and it's gonna stop, continue or exit, so
1458 * there always is something to wait for.
1460 wo->notask_error = 0;
1464 * Wait for stopped. Depending on @ptrace, different stopped state
1465 * is used and the two don't interact with each other.
1467 ret = wait_task_stopped(wo, ptrace, p);
1472 * Wait for continued. There's only one continued state and the
1473 * ptracer can consume it which can confuse the real parent. Don't
1474 * use WCONTINUED from ptracer. You don't need or want it.
1476 return wait_task_continued(wo, p);
1480 * Do the work of do_wait() for one thread in the group, @tsk.
1482 * -ECHILD should be in ->notask_error before the first call.
1483 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1484 * Returns zero if the search for a child should continue; then
1485 * ->notask_error is 0 if there were any eligible children,
1486 * or another error from security_task_wait(), or still -ECHILD.
1488 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1490 struct task_struct *p;
1492 list_for_each_entry(p, &tsk->children, sibling) {
1493 int ret = wait_consider_task(wo, 0, p);
1502 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1504 struct task_struct *p;
1506 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1507 int ret = wait_consider_task(wo, 1, p);
1516 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1517 int sync, void *key)
1519 struct wait_opts *wo = container_of(wait, struct wait_opts,
1521 struct task_struct *p = key;
1523 if (!eligible_pid(wo, p))
1526 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1529 return default_wake_function(wait, mode, sync, key);
1532 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1534 __wake_up_sync_key(&parent->signal->wait_chldexit,
1535 TASK_INTERRUPTIBLE, 1, p);
1538 static long do_wait(struct wait_opts *wo)
1540 struct task_struct *tsk;
1543 trace_sched_process_wait(wo->wo_pid);
1545 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1546 wo->child_wait.private = current;
1547 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1550 * If there is nothing that can match our criteria, just get out.
1551 * We will clear ->notask_error to zero if we see any child that
1552 * might later match our criteria, even if we are not able to reap
1555 wo->notask_error = -ECHILD;
1556 if ((wo->wo_type < PIDTYPE_MAX) &&
1557 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1560 set_current_state(TASK_INTERRUPTIBLE);
1561 read_lock(&tasklist_lock);
1564 retval = do_wait_thread(wo, tsk);
1568 retval = ptrace_do_wait(wo, tsk);
1572 if (wo->wo_flags & __WNOTHREAD)
1574 } while_each_thread(current, tsk);
1575 read_unlock(&tasklist_lock);
1578 retval = wo->notask_error;
1579 if (!retval && !(wo->wo_flags & WNOHANG)) {
1580 retval = -ERESTARTSYS;
1581 if (!signal_pending(current)) {
1587 __set_current_state(TASK_RUNNING);
1588 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1592 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1593 infop, int, options, struct rusage __user *, ru)
1595 struct wait_opts wo;
1596 struct pid *pid = NULL;
1600 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1601 __WNOTHREAD|__WCLONE|__WALL))
1603 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1616 type = PIDTYPE_PGID;
1624 if (type < PIDTYPE_MAX)
1625 pid = find_get_pid(upid);
1629 wo.wo_flags = options;
1639 * For a WNOHANG return, clear out all the fields
1640 * we would set so the user can easily tell the
1644 ret = put_user(0, &infop->si_signo);
1646 ret = put_user(0, &infop->si_errno);
1648 ret = put_user(0, &infop->si_code);
1650 ret = put_user(0, &infop->si_pid);
1652 ret = put_user(0, &infop->si_uid);
1654 ret = put_user(0, &infop->si_status);
1661 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1662 int, options, struct rusage __user *, ru)
1664 struct wait_opts wo;
1665 struct pid *pid = NULL;
1669 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1670 __WNOTHREAD|__WCLONE|__WALL))
1673 /* -INT_MIN is not defined */
1674 if (upid == INT_MIN)
1679 else if (upid < 0) {
1680 type = PIDTYPE_PGID;
1681 pid = find_get_pid(-upid);
1682 } else if (upid == 0) {
1683 type = PIDTYPE_PGID;
1684 pid = get_task_pid(current, PIDTYPE_PGID);
1685 } else /* upid > 0 */ {
1687 pid = find_get_pid(upid);
1692 wo.wo_flags = options | WEXITED;
1694 wo.wo_stat = stat_addr;
1702 #ifdef __ARCH_WANT_SYS_WAITPID
1705 * sys_waitpid() remains for compatibility. waitpid() should be
1706 * implemented by calling sys_wait4() from libc.a.
1708 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1710 return sys_wait4(pid, stat_addr, options, NULL);