4 * Copyright (C) 1991, 1992 Linus Torvalds
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>
66 #include <linux/uaccess.h>
67 #include <asm/unistd.h>
68 #include <asm/pgtable.h>
69 #include <asm/mmu_context.h>
71 static void __unhash_process(struct task_struct *p, bool group_dead)
74 detach_pid(p, PIDTYPE_PID);
76 detach_pid(p, PIDTYPE_PGID);
77 detach_pid(p, PIDTYPE_SID);
79 list_del_rcu(&p->tasks);
80 list_del_init(&p->sibling);
81 __this_cpu_dec(process_counts);
83 list_del_rcu(&p->thread_group);
84 list_del_rcu(&p->thread_node);
88 * This function expects the tasklist_lock write-locked.
90 static void __exit_signal(struct task_struct *tsk)
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);
98 sighand = rcu_dereference_check(tsk->sighand,
99 lockdep_tasklist_lock_is_held());
100 spin_lock(&sighand->siglock);
102 #ifdef CONFIG_POSIX_TIMERS
103 posix_cpu_timers_exit(tsk);
105 posix_cpu_timers_exit_group(tsk);
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.
112 if (unlikely(has_group_leader_pid(tsk)))
113 posix_cpu_timers_exit_group(tsk);
122 * If there is any task waiting for the group exit
125 if (sig->notify_count > 0 && !--sig->notify_count)
126 wake_up_process(sig->group_exit_task);
128 if (tsk == sig->curr_target)
129 sig->curr_target = next_thread(tsk);
132 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
133 sizeof(unsigned long long));
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.
141 task_cputime(tsk, &utime, &stime);
142 write_seqlock(&sig->stats_lock);
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;
155 __unhash_process(tsk, group_dead);
156 write_sequnlock(&sig->stats_lock);
159 * Do this under ->siglock, we can race with another thread
160 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
162 flush_sigqueue(&tsk->pending);
164 spin_unlock(&sighand->siglock);
166 __cleanup_sighand(sighand);
167 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
169 flush_sigqueue(&sig->shared_pending);
174 static void delayed_put_task_struct(struct rcu_head *rhp)
176 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
178 perf_event_delayed_put(tsk);
179 trace_sched_process_free(tsk);
180 put_task_struct(tsk);
184 void release_task(struct task_struct *p)
186 struct task_struct *leader;
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 */
192 atomic_dec(&__task_cred(p)->user->processes);
197 write_lock_irq(&tasklist_lock);
198 ptrace_release_task(p);
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.)
207 leader = p->group_leader;
208 if (leader != p && thread_group_empty(leader)
209 && leader->exit_state == EXIT_ZOMBIE) {
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.
215 zap_leader = do_notify_parent(leader, leader->exit_signal);
217 leader->exit_state = EXIT_DEAD;
220 write_unlock_irq(&tasklist_lock);
222 call_rcu(&p->rcu, delayed_put_task_struct);
225 if (unlikely(zap_leader))
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.
233 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
235 struct sighand_struct *sighand;
236 struct task_struct *task;
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.
245 task = rcu_dereference(*ptask);
249 probe_kernel_address(&task->sighand, sighand);
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
257 if (unlikely(task != READ_ONCE(*ptask)))
261 * We've re-checked that "task == *ptask", now we have two different
264 * 1. This is actually the same task/task_struct. In this case
265 * sighand != NULL tells us it is still alive.
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
271 * In this case we actually return a random value, but this is
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.
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
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.
292 void rcuwait_wake_up(struct rcuwait *w)
294 struct task_struct *task;
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().
305 * [S] tsk = current [S] cond = true
312 * Avoid using task_rcu_dereference() magic as long as we are careful,
313 * see comment in rcuwait_wait_event() regarding ->exit_state.
315 task = rcu_dereference(w->task);
317 wake_up_process(task);
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.
327 * "I ask you, have you ever known what it is to be an orphan?"
329 static int will_become_orphaned_pgrp(struct pid *pgrp,
330 struct task_struct *ignored_task)
332 struct task_struct *p;
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))
340 if (task_pgrp(p->real_parent) != pgrp &&
341 task_session(p->real_parent) == task_session(p))
343 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
348 int is_current_pgrp_orphaned(void)
352 read_lock(&tasklist_lock);
353 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
354 read_unlock(&tasklist_lock);
359 static bool has_stopped_jobs(struct pid *pgrp)
361 struct task_struct *p;
363 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
364 if (p->signal->flags & SIGNAL_STOP_STOPPED)
366 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
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)
377 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
379 struct pid *pgrp = task_pgrp(tsk);
380 struct task_struct *ignored_task = tsk;
383 /* exit: our father is in a different pgrp than
384 * we are and we were the only connection outside.
386 parent = tsk->real_parent;
388 /* reparent: our child is in a different pgrp than
389 * we are, and it was the only connection outside.
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);
404 * A task is exiting. If it owned this mm, find a new owner for the mm.
406 void mm_update_next_owner(struct mm_struct *mm)
408 struct task_struct *c, *g, *p = current;
412 * If the exiting or execing task is not the owner, it's
413 * someone else's problem.
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.
422 if (atomic_read(&mm->mm_users) <= 1) {
427 read_lock(&tasklist_lock);
429 * Search in the children
431 list_for_each_entry(c, &p->children, sibling) {
433 goto assign_new_owner;
437 * Search in the siblings
439 list_for_each_entry(c, &p->real_parent->children, sibling) {
441 goto assign_new_owner;
445 * Search through everything else, we should not get here often.
447 for_each_process(g) {
448 if (g->flags & PF_KTHREAD)
450 for_each_thread(g, c) {
452 goto assign_new_owner;
457 read_unlock(&tasklist_lock);
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.
470 * The task_lock protects c->mm from changing.
471 * We always want mm->owner->mm == mm
475 * Delay read_unlock() till we have the task_lock()
476 * to ensure that c does not slip away underneath us
478 read_unlock(&tasklist_lock);
488 #endif /* CONFIG_MEMCG */
491 * Turn us into a lazy TLB process if we
494 static void exit_mm(void)
496 struct mm_struct *mm = current->mm;
497 struct core_state *core_state;
499 mm_release(current, mm);
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.
510 down_read(&mm->mmap_sem);
511 core_state = mm->core_state;
513 struct core_thread self;
515 up_read(&mm->mmap_sem);
518 self.next = xchg(&core_state->dumper.next, &self);
520 * Implies mb(), the result of xchg() must be visible
521 * to core_state->dumper.
523 if (atomic_dec_and_test(&core_state->nr_threads))
524 complete(&core_state->startup);
527 set_current_state(TASK_UNINTERRUPTIBLE);
528 if (!self.task) /* see coredump_finish() */
530 freezable_schedule();
532 __set_current_state(TASK_RUNNING);
533 down_read(&mm->mmap_sem);
536 BUG_ON(mm != current->active_mm);
537 /* more a memory barrier than a real lock */
540 up_read(&mm->mmap_sem);
541 enter_lazy_tlb(mm, current);
542 task_unlock(current);
543 mm_update_next_owner(mm);
545 if (test_thread_flag(TIF_MEMDIE))
549 static struct task_struct *find_alive_thread(struct task_struct *p)
551 struct task_struct *t;
553 for_each_thread(p, t) {
554 if (!(t->flags & PF_EXITING))
560 static struct task_struct *find_child_reaper(struct task_struct *father,
561 struct list_head *dead)
562 __releases(&tasklist_lock)
563 __acquires(&tasklist_lock)
565 struct pid_namespace *pid_ns = task_active_pid_ns(father);
566 struct task_struct *reaper = pid_ns->child_reaper;
567 struct task_struct *p, *n;
569 if (likely(reaper != father))
572 reaper = find_alive_thread(father);
574 pid_ns->child_reaper = reaper;
578 write_unlock_irq(&tasklist_lock);
579 if (unlikely(pid_ns == &init_pid_ns)) {
580 panic("Attempted to kill init! exitcode=0x%08x\n",
581 father->signal->group_exit_code ?: father->exit_code);
584 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
585 list_del_init(&p->ptrace_entry);
589 zap_pid_ns_processes(pid_ns);
590 write_lock_irq(&tasklist_lock);
596 * When we die, we re-parent all our children, and try to:
597 * 1. give them to another thread in our thread group, if such a member exists
598 * 2. give it to the first ancestor process which prctl'd itself as a
599 * child_subreaper for its children (like a service manager)
600 * 3. give it to the init process (PID 1) in our pid namespace
602 static struct task_struct *find_new_reaper(struct task_struct *father,
603 struct task_struct *child_reaper)
605 struct task_struct *thread, *reaper;
607 thread = find_alive_thread(father);
611 if (father->signal->has_child_subreaper) {
612 unsigned int ns_level = task_pid(father)->level;
614 * Find the first ->is_child_subreaper ancestor in our pid_ns.
615 * We can't check reaper != child_reaper to ensure we do not
616 * cross the namespaces, the exiting parent could be injected
617 * by setns() + fork().
618 * We check pid->level, this is slightly more efficient than
619 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
621 for (reaper = father->real_parent;
622 task_pid(reaper)->level == ns_level;
623 reaper = reaper->real_parent) {
624 if (reaper == &init_task)
626 if (!reaper->signal->is_child_subreaper)
628 thread = find_alive_thread(reaper);
638 * Any that need to be release_task'd are put on the @dead list.
640 static void reparent_leader(struct task_struct *father, struct task_struct *p,
641 struct list_head *dead)
643 if (unlikely(p->exit_state == EXIT_DEAD))
646 /* We don't want people slaying init. */
647 p->exit_signal = SIGCHLD;
649 /* If it has exited notify the new parent about this child's death. */
651 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
652 if (do_notify_parent(p, p->exit_signal)) {
653 p->exit_state = EXIT_DEAD;
654 list_add(&p->ptrace_entry, dead);
658 kill_orphaned_pgrp(p, father);
662 * This does two things:
664 * A. Make init inherit all the child processes
665 * B. Check to see if any process groups have become orphaned
666 * as a result of our exiting, and if they have any stopped
667 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
669 static void forget_original_parent(struct task_struct *father,
670 struct list_head *dead)
672 struct task_struct *p, *t, *reaper;
674 if (unlikely(!list_empty(&father->ptraced)))
675 exit_ptrace(father, dead);
677 /* Can drop and reacquire tasklist_lock */
678 reaper = find_child_reaper(father, dead);
679 if (list_empty(&father->children))
682 reaper = find_new_reaper(father, reaper);
683 list_for_each_entry(p, &father->children, sibling) {
684 for_each_thread(p, t) {
685 t->real_parent = reaper;
686 BUG_ON((!t->ptrace) != (t->parent == father));
687 if (likely(!t->ptrace))
688 t->parent = t->real_parent;
689 if (t->pdeath_signal)
690 group_send_sig_info(t->pdeath_signal,
694 * If this is a threaded reparent there is no need to
695 * notify anyone anything has happened.
697 if (!same_thread_group(reaper, father))
698 reparent_leader(father, p, dead);
700 list_splice_tail_init(&father->children, &reaper->children);
704 * Send signals to all our closest relatives so that they know
705 * to properly mourn us..
707 static void exit_notify(struct task_struct *tsk, int group_dead)
710 struct task_struct *p, *n;
713 write_lock_irq(&tasklist_lock);
714 forget_original_parent(tsk, &dead);
717 kill_orphaned_pgrp(tsk->group_leader, NULL);
719 if (unlikely(tsk->ptrace)) {
720 int sig = thread_group_leader(tsk) &&
721 thread_group_empty(tsk) &&
722 !ptrace_reparented(tsk) ?
723 tsk->exit_signal : SIGCHLD;
724 autoreap = do_notify_parent(tsk, sig);
725 } else if (thread_group_leader(tsk)) {
726 autoreap = thread_group_empty(tsk) &&
727 do_notify_parent(tsk, tsk->exit_signal);
732 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
733 if (tsk->exit_state == EXIT_DEAD)
734 list_add(&tsk->ptrace_entry, &dead);
736 /* mt-exec, de_thread() is waiting for group leader */
737 if (unlikely(tsk->signal->notify_count < 0))
738 wake_up_process(tsk->signal->group_exit_task);
739 write_unlock_irq(&tasklist_lock);
741 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
742 list_del_init(&p->ptrace_entry);
747 #ifdef CONFIG_DEBUG_STACK_USAGE
748 static void check_stack_usage(void)
750 static DEFINE_SPINLOCK(low_water_lock);
751 static int lowest_to_date = THREAD_SIZE;
754 free = stack_not_used(current);
756 if (free >= lowest_to_date)
759 spin_lock(&low_water_lock);
760 if (free < lowest_to_date) {
761 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
762 current->comm, task_pid_nr(current), free);
763 lowest_to_date = free;
765 spin_unlock(&low_water_lock);
768 static inline void check_stack_usage(void) {}
771 void __noreturn do_exit(long code)
773 struct task_struct *tsk = current;
776 profile_task_exit(tsk);
779 WARN_ON(blk_needs_flush_plug(tsk));
781 if (unlikely(in_interrupt()))
782 panic("Aiee, killing interrupt handler!");
783 if (unlikely(!tsk->pid))
784 panic("Attempted to kill the idle task!");
787 * If do_exit is called because this processes oopsed, it's possible
788 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
789 * continuing. Amongst other possible reasons, this is to prevent
790 * mm_release()->clear_child_tid() from writing to a user-controlled
795 ptrace_event(PTRACE_EVENT_EXIT, code);
797 validate_creds_for_do_exit(tsk);
800 * We're taking recursive faults here in do_exit. Safest is to just
801 * leave this task alone and wait for reboot.
803 if (unlikely(tsk->flags & PF_EXITING)) {
804 pr_alert("Fixing recursive fault but reboot is needed!\n");
806 * We can do this unlocked here. The futex code uses
807 * this flag just to verify whether the pi state
808 * cleanup has been done or not. In the worst case it
809 * loops once more. We pretend that the cleanup was
810 * done as there is no way to return. Either the
811 * OWNER_DIED bit is set by now or we push the blocked
812 * task into the wait for ever nirwana as well.
814 tsk->flags |= PF_EXITPIDONE;
815 set_current_state(TASK_UNINTERRUPTIBLE);
819 exit_signals(tsk); /* sets PF_EXITING */
821 * Ensure that all new tsk->pi_lock acquisitions must observe
822 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
826 * Ensure that we must observe the pi_state in exit_mm() ->
827 * mm_release() -> exit_pi_state_list().
829 raw_spin_lock_irq(&tsk->pi_lock);
830 raw_spin_unlock_irq(&tsk->pi_lock);
832 if (unlikely(in_atomic())) {
833 pr_info("note: %s[%d] exited with preempt_count %d\n",
834 current->comm, task_pid_nr(current),
836 preempt_count_set(PREEMPT_ENABLED);
839 /* sync mm's RSS info before statistics gathering */
841 sync_mm_rss(tsk->mm);
842 acct_update_integrals(tsk);
843 group_dead = atomic_dec_and_test(&tsk->signal->live);
845 #ifdef CONFIG_POSIX_TIMERS
846 hrtimer_cancel(&tsk->signal->real_timer);
847 exit_itimers(tsk->signal);
850 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
852 acct_collect(code, group_dead);
857 tsk->exit_code = code;
858 taskstats_exit(tsk, group_dead);
864 trace_sched_process_exit(tsk);
871 disassociate_ctty(1);
872 exit_task_namespaces(tsk);
877 * Flush inherited counters to the parent - before the parent
878 * gets woken up by child-exit notifications.
880 * because of cgroup mode, must be called before cgroup_exit()
882 perf_event_exit_task(tsk);
884 sched_autogroup_exit_task(tsk);
888 * FIXME: do that only when needed, using sched_exit tracepoint
890 flush_ptrace_hw_breakpoint(tsk);
892 exit_tasks_rcu_start();
893 exit_notify(tsk, group_dead);
894 proc_exit_connector(tsk);
895 mpol_put_task_policy(tsk);
897 if (unlikely(current->pi_state_cache))
898 kfree(current->pi_state_cache);
901 * Make sure we are holding no locks:
903 debug_check_no_locks_held();
905 * We can do this unlocked here. The futex code uses this flag
906 * just to verify whether the pi state cleanup has been done
907 * or not. In the worst case it loops once more.
909 tsk->flags |= PF_EXITPIDONE;
912 exit_io_context(tsk);
914 if (tsk->splice_pipe)
915 free_pipe_info(tsk->splice_pipe);
917 if (tsk->task_frag.page)
918 put_page(tsk->task_frag.page);
920 validate_creds_for_do_exit(tsk);
925 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
927 exit_tasks_rcu_finish();
929 lockdep_free_task(tsk);
932 EXPORT_SYMBOL_GPL(do_exit);
934 void complete_and_exit(struct completion *comp, long code)
941 EXPORT_SYMBOL(complete_and_exit);
943 SYSCALL_DEFINE1(exit, int, error_code)
945 do_exit((error_code&0xff)<<8);
949 * Take down every thread in the group. This is called by fatal signals
950 * as well as by sys_exit_group (below).
953 do_group_exit(int exit_code)
955 struct signal_struct *sig = current->signal;
957 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
959 if (signal_group_exit(sig))
960 exit_code = sig->group_exit_code;
961 else if (!thread_group_empty(current)) {
962 struct sighand_struct *const sighand = current->sighand;
964 spin_lock_irq(&sighand->siglock);
965 if (signal_group_exit(sig))
966 /* Another thread got here before we took the lock. */
967 exit_code = sig->group_exit_code;
969 sig->group_exit_code = exit_code;
970 sig->flags = SIGNAL_GROUP_EXIT;
971 zap_other_threads(current);
973 spin_unlock_irq(&sighand->siglock);
981 * this kills every thread in the thread group. Note that any externally
982 * wait4()-ing process will get the correct exit code - even if this
983 * thread is not the thread group leader.
985 SYSCALL_DEFINE1(exit_group, int, error_code)
987 do_group_exit((error_code & 0xff) << 8);
1000 enum pid_type wo_type;
1004 struct waitid_info *wo_info;
1006 struct rusage *wo_rusage;
1008 wait_queue_entry_t child_wait;
1013 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1015 if (type != PIDTYPE_PID)
1016 task = task->group_leader;
1017 return task->pids[type].pid;
1020 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1022 return wo->wo_type == PIDTYPE_MAX ||
1023 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1027 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1029 if (!eligible_pid(wo, p))
1033 * Wait for all children (clone and not) if __WALL is set or
1034 * if it is traced by us.
1036 if (ptrace || (wo->wo_flags & __WALL))
1040 * Otherwise, wait for clone children *only* if __WCLONE is set;
1041 * otherwise, wait for non-clone children *only*.
1043 * Note: a "clone" child here is one that reports to its parent
1044 * using a signal other than SIGCHLD, or a non-leader thread which
1045 * we can only see if it is traced by us.
1047 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1054 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1055 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1056 * the lock and this task is uninteresting. If we return nonzero, we have
1057 * released the lock and the system call should return.
1059 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1062 pid_t pid = task_pid_vnr(p);
1063 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1064 struct waitid_info *infop;
1066 if (!likely(wo->wo_flags & WEXITED))
1069 if (unlikely(wo->wo_flags & WNOWAIT)) {
1070 status = p->exit_code;
1072 read_unlock(&tasklist_lock);
1073 sched_annotate_sleep();
1075 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1080 * Move the task's state to DEAD/TRACE, only one thread can do this.
1082 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1083 EXIT_TRACE : EXIT_DEAD;
1084 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1087 * We own this thread, nobody else can reap it.
1089 read_unlock(&tasklist_lock);
1090 sched_annotate_sleep();
1093 * Check thread_group_leader() to exclude the traced sub-threads.
1095 if (state == EXIT_DEAD && thread_group_leader(p)) {
1096 struct signal_struct *sig = p->signal;
1097 struct signal_struct *psig = current->signal;
1098 unsigned long maxrss;
1099 u64 tgutime, tgstime;
1102 * The resource counters for the group leader are in its
1103 * own task_struct. Those for dead threads in the group
1104 * are in its signal_struct, as are those for the child
1105 * processes it has previously reaped. All these
1106 * accumulate in the parent's signal_struct c* fields.
1108 * We don't bother to take a lock here to protect these
1109 * p->signal fields because the whole thread group is dead
1110 * and nobody can change them.
1112 * psig->stats_lock also protects us from our sub-theads
1113 * which can reap other children at the same time. Until
1114 * we change k_getrusage()-like users to rely on this lock
1115 * we have to take ->siglock as well.
1117 * We use thread_group_cputime_adjusted() to get times for
1118 * the thread group, which consolidates times for all threads
1119 * in the group including the group leader.
1121 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1122 spin_lock_irq(¤t->sighand->siglock);
1123 write_seqlock(&psig->stats_lock);
1124 psig->cutime += tgutime + sig->cutime;
1125 psig->cstime += tgstime + sig->cstime;
1126 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1128 p->min_flt + sig->min_flt + sig->cmin_flt;
1130 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1132 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1134 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1136 task_io_get_inblock(p) +
1137 sig->inblock + sig->cinblock;
1139 task_io_get_oublock(p) +
1140 sig->oublock + sig->coublock;
1141 maxrss = max(sig->maxrss, sig->cmaxrss);
1142 if (psig->cmaxrss < maxrss)
1143 psig->cmaxrss = maxrss;
1144 task_io_accounting_add(&psig->ioac, &p->ioac);
1145 task_io_accounting_add(&psig->ioac, &sig->ioac);
1146 write_sequnlock(&psig->stats_lock);
1147 spin_unlock_irq(¤t->sighand->siglock);
1151 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1152 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1153 ? p->signal->group_exit_code : p->exit_code;
1154 wo->wo_stat = status;
1156 if (state == EXIT_TRACE) {
1157 write_lock_irq(&tasklist_lock);
1158 /* We dropped tasklist, ptracer could die and untrace */
1161 /* If parent wants a zombie, don't release it now */
1162 state = EXIT_ZOMBIE;
1163 if (do_notify_parent(p, p->exit_signal))
1165 p->exit_state = state;
1166 write_unlock_irq(&tasklist_lock);
1168 if (state == EXIT_DEAD)
1172 infop = wo->wo_info;
1174 if ((status & 0x7f) == 0) {
1175 infop->cause = CLD_EXITED;
1176 infop->status = status >> 8;
1178 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1179 infop->status = status & 0x7f;
1188 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1191 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1192 return &p->exit_code;
1194 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1195 return &p->signal->group_exit_code;
1201 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1203 * @ptrace: is the wait for ptrace
1204 * @p: task to wait for
1206 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1209 * read_lock(&tasklist_lock), which is released if return value is
1210 * non-zero. Also, grabs and releases @p->sighand->siglock.
1213 * 0 if wait condition didn't exist and search for other wait conditions
1214 * should continue. Non-zero return, -errno on failure and @p's pid on
1215 * success, implies that tasklist_lock is released and wait condition
1216 * search should terminate.
1218 static int wait_task_stopped(struct wait_opts *wo,
1219 int ptrace, struct task_struct *p)
1221 struct waitid_info *infop;
1222 int exit_code, *p_code, why;
1223 uid_t uid = 0; /* unneeded, required by compiler */
1227 * Traditionally we see ptrace'd stopped tasks regardless of options.
1229 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1232 if (!task_stopped_code(p, ptrace))
1236 spin_lock_irq(&p->sighand->siglock);
1238 p_code = task_stopped_code(p, ptrace);
1239 if (unlikely(!p_code))
1242 exit_code = *p_code;
1246 if (!unlikely(wo->wo_flags & WNOWAIT))
1249 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1251 spin_unlock_irq(&p->sighand->siglock);
1256 * Now we are pretty sure this task is interesting.
1257 * Make sure it doesn't get reaped out from under us while we
1258 * give up the lock and then examine it below. We don't want to
1259 * keep holding onto the tasklist_lock while we call getrusage and
1260 * possibly take page faults for user memory.
1263 pid = task_pid_vnr(p);
1264 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1265 read_unlock(&tasklist_lock);
1266 sched_annotate_sleep();
1268 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1271 if (likely(!(wo->wo_flags & WNOWAIT)))
1272 wo->wo_stat = (exit_code << 8) | 0x7f;
1274 infop = wo->wo_info;
1277 infop->status = exit_code;
1285 * Handle do_wait work for one task in a live, non-stopped state.
1286 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1287 * the lock and this task is uninteresting. If we return nonzero, we have
1288 * released the lock and the system call should return.
1290 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1292 struct waitid_info *infop;
1296 if (!unlikely(wo->wo_flags & WCONTINUED))
1299 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1302 spin_lock_irq(&p->sighand->siglock);
1303 /* Re-check with the lock held. */
1304 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1305 spin_unlock_irq(&p->sighand->siglock);
1308 if (!unlikely(wo->wo_flags & WNOWAIT))
1309 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1310 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1311 spin_unlock_irq(&p->sighand->siglock);
1313 pid = task_pid_vnr(p);
1315 read_unlock(&tasklist_lock);
1316 sched_annotate_sleep();
1318 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1321 infop = wo->wo_info;
1323 wo->wo_stat = 0xffff;
1325 infop->cause = CLD_CONTINUED;
1328 infop->status = SIGCONT;
1334 * Consider @p for a wait by @parent.
1336 * -ECHILD should be in ->notask_error before the first call.
1337 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1338 * Returns zero if the search for a child should continue;
1339 * then ->notask_error is 0 if @p is an eligible child,
1342 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1343 struct task_struct *p)
1346 * We can race with wait_task_zombie() from another thread.
1347 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1348 * can't confuse the checks below.
1350 int exit_state = ACCESS_ONCE(p->exit_state);
1353 if (unlikely(exit_state == EXIT_DEAD))
1356 ret = eligible_child(wo, ptrace, p);
1360 if (unlikely(exit_state == EXIT_TRACE)) {
1362 * ptrace == 0 means we are the natural parent. In this case
1363 * we should clear notask_error, debugger will notify us.
1365 if (likely(!ptrace))
1366 wo->notask_error = 0;
1370 if (likely(!ptrace) && unlikely(p->ptrace)) {
1372 * If it is traced by its real parent's group, just pretend
1373 * the caller is ptrace_do_wait() and reap this child if it
1376 * This also hides group stop state from real parent; otherwise
1377 * a single stop can be reported twice as group and ptrace stop.
1378 * If a ptracer wants to distinguish these two events for its
1379 * own children it should create a separate process which takes
1380 * the role of real parent.
1382 if (!ptrace_reparented(p))
1387 if (exit_state == EXIT_ZOMBIE) {
1388 /* we don't reap group leaders with subthreads */
1389 if (!delay_group_leader(p)) {
1391 * A zombie ptracee is only visible to its ptracer.
1392 * Notification and reaping will be cascaded to the
1393 * real parent when the ptracer detaches.
1395 if (unlikely(ptrace) || likely(!p->ptrace))
1396 return wait_task_zombie(wo, p);
1400 * Allow access to stopped/continued state via zombie by
1401 * falling through. Clearing of notask_error is complex.
1405 * If WEXITED is set, notask_error should naturally be
1406 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1407 * so, if there are live subthreads, there are events to
1408 * wait for. If all subthreads are dead, it's still safe
1409 * to clear - this function will be called again in finite
1410 * amount time once all the subthreads are released and
1411 * will then return without clearing.
1415 * Stopped state is per-task and thus can't change once the
1416 * target task dies. Only continued and exited can happen.
1417 * Clear notask_error if WCONTINUED | WEXITED.
1419 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1420 wo->notask_error = 0;
1423 * @p is alive and it's gonna stop, continue or exit, so
1424 * there always is something to wait for.
1426 wo->notask_error = 0;
1430 * Wait for stopped. Depending on @ptrace, different stopped state
1431 * is used and the two don't interact with each other.
1433 ret = wait_task_stopped(wo, ptrace, p);
1438 * Wait for continued. There's only one continued state and the
1439 * ptracer can consume it which can confuse the real parent. Don't
1440 * use WCONTINUED from ptracer. You don't need or want it.
1442 return wait_task_continued(wo, p);
1446 * Do the work of do_wait() for one thread in the group, @tsk.
1448 * -ECHILD should be in ->notask_error before the first call.
1449 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1450 * Returns zero if the search for a child should continue; then
1451 * ->notask_error is 0 if there were any eligible children,
1454 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1456 struct task_struct *p;
1458 list_for_each_entry(p, &tsk->children, sibling) {
1459 int ret = wait_consider_task(wo, 0, p);
1468 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1470 struct task_struct *p;
1472 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1473 int ret = wait_consider_task(wo, 1, p);
1482 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1483 int sync, void *key)
1485 struct wait_opts *wo = container_of(wait, struct wait_opts,
1487 struct task_struct *p = key;
1489 if (!eligible_pid(wo, p))
1492 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1495 return default_wake_function(wait, mode, sync, key);
1498 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1500 __wake_up_sync_key(&parent->signal->wait_chldexit,
1501 TASK_INTERRUPTIBLE, 1, p);
1504 static long do_wait(struct wait_opts *wo)
1506 struct task_struct *tsk;
1509 trace_sched_process_wait(wo->wo_pid);
1511 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1512 wo->child_wait.private = current;
1513 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1516 * If there is nothing that can match our criteria, just get out.
1517 * We will clear ->notask_error to zero if we see any child that
1518 * might later match our criteria, even if we are not able to reap
1521 wo->notask_error = -ECHILD;
1522 if ((wo->wo_type < PIDTYPE_MAX) &&
1523 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1526 set_current_state(TASK_INTERRUPTIBLE);
1527 read_lock(&tasklist_lock);
1530 retval = do_wait_thread(wo, tsk);
1534 retval = ptrace_do_wait(wo, tsk);
1538 if (wo->wo_flags & __WNOTHREAD)
1540 } while_each_thread(current, tsk);
1541 read_unlock(&tasklist_lock);
1544 retval = wo->notask_error;
1545 if (!retval && !(wo->wo_flags & WNOHANG)) {
1546 retval = -ERESTARTSYS;
1547 if (!signal_pending(current)) {
1553 __set_current_state(TASK_RUNNING);
1554 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1558 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1559 int options, struct rusage *ru)
1561 struct wait_opts wo;
1562 struct pid *pid = NULL;
1566 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1567 __WNOTHREAD|__WCLONE|__WALL))
1569 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1582 type = PIDTYPE_PGID;
1590 if (type < PIDTYPE_MAX)
1591 pid = find_get_pid(upid);
1595 wo.wo_flags = options;
1604 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1605 infop, int, options, struct rusage __user *, ru)
1608 struct waitid_info info = {.status = 0};
1609 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1615 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1621 if (!access_ok(VERIFY_WRITE, infop, sizeof(*infop)))
1624 user_access_begin();
1625 unsafe_put_user(signo, &infop->si_signo, Efault);
1626 unsafe_put_user(0, &infop->si_errno, Efault);
1627 unsafe_put_user(info.cause, &infop->si_code, Efault);
1628 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1629 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1630 unsafe_put_user(info.status, &infop->si_status, Efault);
1638 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1641 struct wait_opts wo;
1642 struct pid *pid = NULL;
1646 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1647 __WNOTHREAD|__WCLONE|__WALL))
1650 /* -INT_MIN is not defined */
1651 if (upid == INT_MIN)
1656 else if (upid < 0) {
1657 type = PIDTYPE_PGID;
1658 pid = find_get_pid(-upid);
1659 } else if (upid == 0) {
1660 type = PIDTYPE_PGID;
1661 pid = get_task_pid(current, PIDTYPE_PGID);
1662 } else /* upid > 0 */ {
1664 pid = find_get_pid(upid);
1669 wo.wo_flags = options | WEXITED;
1675 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1681 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1682 int, options, struct rusage __user *, ru)
1685 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1688 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1694 #ifdef __ARCH_WANT_SYS_WAITPID
1697 * sys_waitpid() remains for compatibility. waitpid() should be
1698 * implemented by calling sys_wait4() from libc.a.
1700 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1702 return sys_wait4(pid, stat_addr, options, NULL);
1707 #ifdef CONFIG_COMPAT
1708 COMPAT_SYSCALL_DEFINE4(wait4,
1710 compat_uint_t __user *, stat_addr,
1712 struct compat_rusage __user *, ru)
1715 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1717 if (ru && put_compat_rusage(&r, ru))
1723 COMPAT_SYSCALL_DEFINE5(waitid,
1724 int, which, compat_pid_t, pid,
1725 struct compat_siginfo __user *, infop, int, options,
1726 struct compat_rusage __user *, uru)
1729 struct waitid_info info = {.status = 0};
1730 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1736 /* kernel_waitid() overwrites everything in ru */
1737 if (COMPAT_USE_64BIT_TIME)
1738 err = copy_to_user(uru, &ru, sizeof(ru));
1740 err = put_compat_rusage(&ru, uru);
1749 if (!access_ok(VERIFY_WRITE, infop, sizeof(*infop)))
1752 user_access_begin();
1753 unsafe_put_user(signo, &infop->si_signo, Efault);
1754 unsafe_put_user(0, &infop->si_errno, Efault);
1755 unsafe_put_user(info.cause, &infop->si_code, Efault);
1756 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1757 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1758 unsafe_put_user(info.status, &infop->si_status, Efault);
1767 __weak void abort(void)
1771 /* if that doesn't kill us, halt */
1772 panic("Oops failed to kill thread");
1774 EXPORT_SYMBOL(abort);