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/userfaultfd_k.h>
55 #include <linux/init_task.h>
56 #include <linux/perf_event.h>
57 #include <trace/events/sched.h>
58 #include <linux/hw_breakpoint.h>
59 #include <linux/oom.h>
60 #include <linux/writeback.h>
61 #include <linux/shm.h>
62 #include <linux/kcov.h>
63 #include <linux/random.h>
64 #include <linux/rcuwait.h>
65 #include <linux/compat.h>
67 #include <linux/uaccess.h>
68 #include <asm/unistd.h>
69 #include <asm/pgtable.h>
70 #include <asm/mmu_context.h>
72 static void __unhash_process(struct task_struct *p, bool group_dead)
75 detach_pid(p, PIDTYPE_PID);
77 detach_pid(p, PIDTYPE_PGID);
78 detach_pid(p, PIDTYPE_SID);
80 list_del_rcu(&p->tasks);
81 list_del_init(&p->sibling);
82 __this_cpu_dec(process_counts);
84 list_del_rcu(&p->thread_group);
85 list_del_rcu(&p->thread_node);
89 * This function expects the tasklist_lock write-locked.
91 static void __exit_signal(struct task_struct *tsk)
93 struct signal_struct *sig = tsk->signal;
94 bool group_dead = thread_group_leader(tsk);
95 struct sighand_struct *sighand;
96 struct tty_struct *uninitialized_var(tty);
99 sighand = rcu_dereference_check(tsk->sighand,
100 lockdep_tasklist_lock_is_held());
101 spin_lock(&sighand->siglock);
103 #ifdef CONFIG_POSIX_TIMERS
104 posix_cpu_timers_exit(tsk);
106 posix_cpu_timers_exit_group(tsk);
109 * This can only happen if the caller is de_thread().
110 * FIXME: this is the temporary hack, we should teach
111 * posix-cpu-timers to handle this case correctly.
113 if (unlikely(has_group_leader_pid(tsk)))
114 posix_cpu_timers_exit_group(tsk);
123 * If there is any task waiting for the group exit
126 if (sig->notify_count > 0 && !--sig->notify_count)
127 wake_up_process(sig->group_exit_task);
129 if (tsk == sig->curr_target)
130 sig->curr_target = next_thread(tsk);
133 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
134 sizeof(unsigned long long));
137 * Accumulate here the counters for all threads as they die. We could
138 * skip the group leader because it is the last user of signal_struct,
139 * but we want to avoid the race with thread_group_cputime() which can
140 * see the empty ->thread_head list.
142 task_cputime(tsk, &utime, &stime);
143 write_seqlock(&sig->stats_lock);
146 sig->gtime += task_gtime(tsk);
147 sig->min_flt += tsk->min_flt;
148 sig->maj_flt += tsk->maj_flt;
149 sig->nvcsw += tsk->nvcsw;
150 sig->nivcsw += tsk->nivcsw;
151 sig->inblock += task_io_get_inblock(tsk);
152 sig->oublock += task_io_get_oublock(tsk);
153 task_io_accounting_add(&sig->ioac, &tsk->ioac);
154 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
156 __unhash_process(tsk, group_dead);
157 write_sequnlock(&sig->stats_lock);
160 * Do this under ->siglock, we can race with another thread
161 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
163 flush_sigqueue(&tsk->pending);
165 spin_unlock(&sighand->siglock);
167 __cleanup_sighand(sighand);
168 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
170 flush_sigqueue(&sig->shared_pending);
175 static void delayed_put_task_struct(struct rcu_head *rhp)
177 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
179 perf_event_delayed_put(tsk);
180 trace_sched_process_free(tsk);
181 put_task_struct(tsk);
185 void release_task(struct task_struct *p)
187 struct task_struct *leader;
190 /* don't need to get the RCU readlock here - the process is dead and
191 * can't be modifying its own credentials. But shut RCU-lockdep up */
193 atomic_dec(&__task_cred(p)->user->processes);
198 write_lock_irq(&tasklist_lock);
199 ptrace_release_task(p);
203 * If we are the last non-leader member of the thread
204 * group, and the leader is zombie, then notify the
205 * group leader's parent process. (if it wants notification.)
208 leader = p->group_leader;
209 if (leader != p && thread_group_empty(leader)
210 && leader->exit_state == EXIT_ZOMBIE) {
212 * If we were the last child thread and the leader has
213 * exited already, and the leader's parent ignores SIGCHLD,
214 * then we are the one who should release the leader.
216 zap_leader = do_notify_parent(leader, leader->exit_signal);
218 leader->exit_state = EXIT_DEAD;
221 write_unlock_irq(&tasklist_lock);
223 call_rcu(&p->rcu, delayed_put_task_struct);
226 if (unlikely(zap_leader))
231 * Note that if this function returns a valid task_struct pointer (!NULL)
232 * task->usage must remain >0 for the duration of the RCU critical section.
234 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
236 struct sighand_struct *sighand;
237 struct task_struct *task;
240 * We need to verify that release_task() was not called and thus
241 * delayed_put_task_struct() can't run and drop the last reference
242 * before rcu_read_unlock(). We check task->sighand != NULL,
243 * but we can read the already freed and reused memory.
246 task = rcu_dereference(*ptask);
250 probe_kernel_address(&task->sighand, sighand);
253 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
254 * was already freed we can not miss the preceding update of this
258 if (unlikely(task != READ_ONCE(*ptask)))
262 * We've re-checked that "task == *ptask", now we have two different
265 * 1. This is actually the same task/task_struct. In this case
266 * sighand != NULL tells us it is still alive.
268 * 2. This is another task which got the same memory for task_struct.
269 * We can't know this of course, and we can not trust
272 * In this case we actually return a random value, but this is
275 * If we return NULL - we can pretend that we actually noticed that
276 * *ptask was updated when the previous task has exited. Or pretend
277 * that probe_slab_address(&sighand) reads NULL.
279 * If we return the new task (because sighand is not NULL for any
280 * reason) - this is fine too. This (new) task can't go away before
283 * And note: We could even eliminate the false positive if re-read
284 * task->sighand once again to avoid the falsely NULL. But this case
285 * is very unlikely so we don't care.
293 void rcuwait_wake_up(struct rcuwait *w)
295 struct task_struct *task;
300 * Order condition vs @task, such that everything prior to the load
301 * of @task is visible. This is the condition as to why the user called
302 * rcuwait_trywake() in the first place. Pairs with set_current_state()
303 * barrier (A) in rcuwait_wait_event().
306 * [S] tsk = current [S] cond = true
313 * Avoid using task_rcu_dereference() magic as long as we are careful,
314 * see comment in rcuwait_wait_event() regarding ->exit_state.
316 task = rcu_dereference(w->task);
318 wake_up_process(task);
323 * Determine if a process group is "orphaned", according to the POSIX
324 * definition in 2.2.2.52. Orphaned process groups are not to be affected
325 * by terminal-generated stop signals. Newly orphaned process groups are
326 * to receive a SIGHUP and a SIGCONT.
328 * "I ask you, have you ever known what it is to be an orphan?"
330 static int will_become_orphaned_pgrp(struct pid *pgrp,
331 struct task_struct *ignored_task)
333 struct task_struct *p;
335 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
336 if ((p == ignored_task) ||
337 (p->exit_state && thread_group_empty(p)) ||
338 is_global_init(p->real_parent))
341 if (task_pgrp(p->real_parent) != pgrp &&
342 task_session(p->real_parent) == task_session(p))
344 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
349 int is_current_pgrp_orphaned(void)
353 read_lock(&tasklist_lock);
354 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
355 read_unlock(&tasklist_lock);
360 static bool has_stopped_jobs(struct pid *pgrp)
362 struct task_struct *p;
364 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
365 if (p->signal->flags & SIGNAL_STOP_STOPPED)
367 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
373 * Check to see if any process groups have become orphaned as
374 * a result of our exiting, and if they have any stopped jobs,
375 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
378 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
380 struct pid *pgrp = task_pgrp(tsk);
381 struct task_struct *ignored_task = tsk;
384 /* exit: our father is in a different pgrp than
385 * we are and we were the only connection outside.
387 parent = tsk->real_parent;
389 /* reparent: our child is in a different pgrp than
390 * we are, and it was the only connection outside.
394 if (task_pgrp(parent) != pgrp &&
395 task_session(parent) == task_session(tsk) &&
396 will_become_orphaned_pgrp(pgrp, ignored_task) &&
397 has_stopped_jobs(pgrp)) {
398 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
399 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
405 * A task is exiting. If it owned this mm, find a new owner for the mm.
407 void mm_update_next_owner(struct mm_struct *mm)
409 struct task_struct *c, *g, *p = current;
413 * If the exiting or execing task is not the owner, it's
414 * someone else's problem.
419 * The current owner is exiting/execing and there are no other
420 * candidates. Do not leave the mm pointing to a possibly
421 * freed task structure.
423 if (atomic_read(&mm->mm_users) <= 1) {
428 read_lock(&tasklist_lock);
430 * Search in the children
432 list_for_each_entry(c, &p->children, sibling) {
434 goto assign_new_owner;
438 * Search in the siblings
440 list_for_each_entry(c, &p->real_parent->children, sibling) {
442 goto assign_new_owner;
446 * Search through everything else, we should not get here often.
448 for_each_process(g) {
449 if (g->flags & PF_KTHREAD)
451 for_each_thread(g, c) {
453 goto assign_new_owner;
458 read_unlock(&tasklist_lock);
460 * We found no owner yet mm_users > 1: this implies that we are
461 * most likely racing with swapoff (try_to_unuse()) or /proc or
462 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
471 * The task_lock protects c->mm from changing.
472 * We always want mm->owner->mm == mm
476 * Delay read_unlock() till we have the task_lock()
477 * to ensure that c does not slip away underneath us
479 read_unlock(&tasklist_lock);
489 #endif /* CONFIG_MEMCG */
492 * Turn us into a lazy TLB process if we
495 static void exit_mm(void)
497 struct mm_struct *mm = current->mm;
498 struct core_state *core_state;
500 mm_release(current, mm);
505 * Serialize with any possible pending coredump.
506 * We must hold mmap_sem around checking core_state
507 * and clearing tsk->mm. The core-inducing thread
508 * will increment ->nr_threads for each thread in the
509 * group with ->mm != NULL.
511 down_read(&mm->mmap_sem);
512 core_state = mm->core_state;
514 struct core_thread self;
516 up_read(&mm->mmap_sem);
519 self.next = xchg(&core_state->dumper.next, &self);
521 * Implies mb(), the result of xchg() must be visible
522 * to core_state->dumper.
524 if (atomic_dec_and_test(&core_state->nr_threads))
525 complete(&core_state->startup);
528 set_current_state(TASK_UNINTERRUPTIBLE);
529 if (!self.task) /* see coredump_finish() */
531 freezable_schedule();
533 __set_current_state(TASK_RUNNING);
534 down_read(&mm->mmap_sem);
537 BUG_ON(mm != current->active_mm);
538 /* more a memory barrier than a real lock */
541 up_read(&mm->mmap_sem);
542 enter_lazy_tlb(mm, current);
543 task_unlock(current);
544 mm_update_next_owner(mm);
546 if (test_thread_flag(TIF_MEMDIE))
550 static struct task_struct *find_alive_thread(struct task_struct *p)
552 struct task_struct *t;
554 for_each_thread(p, t) {
555 if (!(t->flags & PF_EXITING))
561 static struct task_struct *find_child_reaper(struct task_struct *father)
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;
568 if (likely(reaper != father))
571 reaper = find_alive_thread(father);
573 pid_ns->child_reaper = reaper;
577 write_unlock_irq(&tasklist_lock);
578 if (unlikely(pid_ns == &init_pid_ns)) {
579 panic("Attempted to kill init! exitcode=0x%08x\n",
580 father->signal->group_exit_code ?: father->exit_code);
582 zap_pid_ns_processes(pid_ns);
583 write_lock_irq(&tasklist_lock);
589 * When we die, we re-parent all our children, and try to:
590 * 1. give them to another thread in our thread group, if such a member exists
591 * 2. give it to the first ancestor process which prctl'd itself as a
592 * child_subreaper for its children (like a service manager)
593 * 3. give it to the init process (PID 1) in our pid namespace
595 static struct task_struct *find_new_reaper(struct task_struct *father,
596 struct task_struct *child_reaper)
598 struct task_struct *thread, *reaper;
600 thread = find_alive_thread(father);
604 if (father->signal->has_child_subreaper) {
605 unsigned int ns_level = task_pid(father)->level;
607 * Find the first ->is_child_subreaper ancestor in our pid_ns.
608 * We can't check reaper != child_reaper to ensure we do not
609 * cross the namespaces, the exiting parent could be injected
610 * by setns() + fork().
611 * We check pid->level, this is slightly more efficient than
612 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
614 for (reaper = father->real_parent;
615 task_pid(reaper)->level == ns_level;
616 reaper = reaper->real_parent) {
617 if (reaper == &init_task)
619 if (!reaper->signal->is_child_subreaper)
621 thread = find_alive_thread(reaper);
631 * Any that need to be release_task'd are put on the @dead list.
633 static void reparent_leader(struct task_struct *father, struct task_struct *p,
634 struct list_head *dead)
636 if (unlikely(p->exit_state == EXIT_DEAD))
639 /* We don't want people slaying init. */
640 p->exit_signal = SIGCHLD;
642 /* If it has exited notify the new parent about this child's death. */
644 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
645 if (do_notify_parent(p, p->exit_signal)) {
646 p->exit_state = EXIT_DEAD;
647 list_add(&p->ptrace_entry, dead);
651 kill_orphaned_pgrp(p, father);
655 * This does two things:
657 * A. Make init inherit all the child processes
658 * B. Check to see if any process groups have become orphaned
659 * as a result of our exiting, and if they have any stopped
660 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
662 static void forget_original_parent(struct task_struct *father,
663 struct list_head *dead)
665 struct task_struct *p, *t, *reaper;
667 if (unlikely(!list_empty(&father->ptraced)))
668 exit_ptrace(father, dead);
670 /* Can drop and reacquire tasklist_lock */
671 reaper = find_child_reaper(father);
672 if (list_empty(&father->children))
675 reaper = find_new_reaper(father, reaper);
676 list_for_each_entry(p, &father->children, sibling) {
677 for_each_thread(p, t) {
678 t->real_parent = reaper;
679 BUG_ON((!t->ptrace) != (t->parent == father));
680 if (likely(!t->ptrace))
681 t->parent = t->real_parent;
682 if (t->pdeath_signal)
683 group_send_sig_info(t->pdeath_signal,
687 * If this is a threaded reparent there is no need to
688 * notify anyone anything has happened.
690 if (!same_thread_group(reaper, father))
691 reparent_leader(father, p, dead);
693 list_splice_tail_init(&father->children, &reaper->children);
697 * Send signals to all our closest relatives so that they know
698 * to properly mourn us..
700 static void exit_notify(struct task_struct *tsk, int group_dead)
703 struct task_struct *p, *n;
706 write_lock_irq(&tasklist_lock);
707 forget_original_parent(tsk, &dead);
710 kill_orphaned_pgrp(tsk->group_leader, NULL);
712 if (unlikely(tsk->ptrace)) {
713 int sig = thread_group_leader(tsk) &&
714 thread_group_empty(tsk) &&
715 !ptrace_reparented(tsk) ?
716 tsk->exit_signal : SIGCHLD;
717 autoreap = do_notify_parent(tsk, sig);
718 } else if (thread_group_leader(tsk)) {
719 autoreap = thread_group_empty(tsk) &&
720 do_notify_parent(tsk, tsk->exit_signal);
725 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
726 if (tsk->exit_state == EXIT_DEAD)
727 list_add(&tsk->ptrace_entry, &dead);
729 /* mt-exec, de_thread() is waiting for group leader */
730 if (unlikely(tsk->signal->notify_count < 0))
731 wake_up_process(tsk->signal->group_exit_task);
732 write_unlock_irq(&tasklist_lock);
734 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
735 list_del_init(&p->ptrace_entry);
740 #ifdef CONFIG_DEBUG_STACK_USAGE
741 static void check_stack_usage(void)
743 static DEFINE_SPINLOCK(low_water_lock);
744 static int lowest_to_date = THREAD_SIZE;
747 free = stack_not_used(current);
749 if (free >= lowest_to_date)
752 spin_lock(&low_water_lock);
753 if (free < lowest_to_date) {
754 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
755 current->comm, task_pid_nr(current), free);
756 lowest_to_date = free;
758 spin_unlock(&low_water_lock);
761 static inline void check_stack_usage(void) {}
764 void __noreturn do_exit(long code)
766 struct task_struct *tsk = current;
768 TASKS_RCU(int tasks_rcu_i);
770 profile_task_exit(tsk);
773 WARN_ON(blk_needs_flush_plug(tsk));
775 if (unlikely(in_interrupt()))
776 panic("Aiee, killing interrupt handler!");
777 if (unlikely(!tsk->pid))
778 panic("Attempted to kill the idle task!");
781 * If do_exit is called because this processes oopsed, it's possible
782 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
783 * continuing. Amongst other possible reasons, this is to prevent
784 * mm_release()->clear_child_tid() from writing to a user-controlled
789 ptrace_event(PTRACE_EVENT_EXIT, code);
791 validate_creds_for_do_exit(tsk);
794 * We're taking recursive faults here in do_exit. Safest is to just
795 * leave this task alone and wait for reboot.
797 if (unlikely(tsk->flags & PF_EXITING)) {
798 pr_alert("Fixing recursive fault but reboot is needed!\n");
800 * We can do this unlocked here. The futex code uses
801 * this flag just to verify whether the pi state
802 * cleanup has been done or not. In the worst case it
803 * loops once more. We pretend that the cleanup was
804 * done as there is no way to return. Either the
805 * OWNER_DIED bit is set by now or we push the blocked
806 * task into the wait for ever nirwana as well.
808 tsk->flags |= PF_EXITPIDONE;
809 set_current_state(TASK_UNINTERRUPTIBLE);
813 exit_signals(tsk); /* sets PF_EXITING */
815 * Ensure that all new tsk->pi_lock acquisitions must observe
816 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
820 * Ensure that we must observe the pi_state in exit_mm() ->
821 * mm_release() -> exit_pi_state_list().
823 raw_spin_unlock_wait(&tsk->pi_lock);
825 if (unlikely(in_atomic())) {
826 pr_info("note: %s[%d] exited with preempt_count %d\n",
827 current->comm, task_pid_nr(current),
829 preempt_count_set(PREEMPT_ENABLED);
832 /* sync mm's RSS info before statistics gathering */
834 sync_mm_rss(tsk->mm);
835 acct_update_integrals(tsk);
836 group_dead = atomic_dec_and_test(&tsk->signal->live);
838 #ifdef CONFIG_POSIX_TIMERS
839 hrtimer_cancel(&tsk->signal->real_timer);
840 exit_itimers(tsk->signal);
843 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
845 acct_collect(code, group_dead);
850 tsk->exit_code = code;
851 taskstats_exit(tsk, group_dead);
857 trace_sched_process_exit(tsk);
864 disassociate_ctty(1);
865 exit_task_namespaces(tsk);
870 * Flush inherited counters to the parent - before the parent
871 * gets woken up by child-exit notifications.
873 * because of cgroup mode, must be called before cgroup_exit()
875 perf_event_exit_task(tsk);
877 sched_autogroup_exit_task(tsk);
881 * FIXME: do that only when needed, using sched_exit tracepoint
883 flush_ptrace_hw_breakpoint(tsk);
885 TASKS_RCU(preempt_disable());
886 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
887 TASKS_RCU(preempt_enable());
888 exit_notify(tsk, group_dead);
889 proc_exit_connector(tsk);
890 mpol_put_task_policy(tsk);
892 if (unlikely(current->pi_state_cache))
893 kfree(current->pi_state_cache);
896 * Make sure we are holding no locks:
898 debug_check_no_locks_held();
900 * We can do this unlocked here. The futex code uses this flag
901 * just to verify whether the pi state cleanup has been done
902 * or not. In the worst case it loops once more.
904 tsk->flags |= PF_EXITPIDONE;
907 exit_io_context(tsk);
909 if (tsk->splice_pipe)
910 free_pipe_info(tsk->splice_pipe);
912 if (tsk->task_frag.page)
913 put_page(tsk->task_frag.page);
915 validate_creds_for_do_exit(tsk);
920 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
922 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
926 EXPORT_SYMBOL_GPL(do_exit);
928 void complete_and_exit(struct completion *comp, long code)
935 EXPORT_SYMBOL(complete_and_exit);
937 SYSCALL_DEFINE1(exit, int, error_code)
939 do_exit((error_code&0xff)<<8);
943 * Take down every thread in the group. This is called by fatal signals
944 * as well as by sys_exit_group (below).
947 do_group_exit(int exit_code)
949 struct signal_struct *sig = current->signal;
951 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
953 if (signal_group_exit(sig))
954 exit_code = sig->group_exit_code;
955 else if (!thread_group_empty(current)) {
956 struct sighand_struct *const sighand = current->sighand;
958 spin_lock_irq(&sighand->siglock);
959 if (signal_group_exit(sig))
960 /* Another thread got here before we took the lock. */
961 exit_code = sig->group_exit_code;
963 sig->group_exit_code = exit_code;
964 sig->flags = SIGNAL_GROUP_EXIT;
965 zap_other_threads(current);
967 spin_unlock_irq(&sighand->siglock);
975 * this kills every thread in the thread group. Note that any externally
976 * wait4()-ing process will get the correct exit code - even if this
977 * thread is not the thread group leader.
979 SYSCALL_DEFINE1(exit_group, int, error_code)
981 do_group_exit((error_code & 0xff) << 8);
994 enum pid_type wo_type;
998 struct waitid_info *wo_info;
1000 struct rusage *wo_rusage;
1002 wait_queue_entry_t child_wait;
1007 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1009 if (type != PIDTYPE_PID)
1010 task = task->group_leader;
1011 return task->pids[type].pid;
1014 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1016 return wo->wo_type == PIDTYPE_MAX ||
1017 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1021 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1023 if (!eligible_pid(wo, p))
1027 * Wait for all children (clone and not) if __WALL is set or
1028 * if it is traced by us.
1030 if (ptrace || (wo->wo_flags & __WALL))
1034 * Otherwise, wait for clone children *only* if __WCLONE is set;
1035 * otherwise, wait for non-clone children *only*.
1037 * Note: a "clone" child here is one that reports to its parent
1038 * using a signal other than SIGCHLD, or a non-leader thread which
1039 * we can only see if it is traced by us.
1041 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1048 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1049 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1050 * the lock and this task is uninteresting. If we return nonzero, we have
1051 * released the lock and the system call should return.
1053 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1056 pid_t pid = task_pid_vnr(p);
1057 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1058 struct waitid_info *infop;
1060 if (!likely(wo->wo_flags & WEXITED))
1063 if (unlikely(wo->wo_flags & WNOWAIT)) {
1064 status = p->exit_code;
1066 read_unlock(&tasklist_lock);
1067 sched_annotate_sleep();
1069 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1074 * Move the task's state to DEAD/TRACE, only one thread can do this.
1076 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1077 EXIT_TRACE : EXIT_DEAD;
1078 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1081 * We own this thread, nobody else can reap it.
1083 read_unlock(&tasklist_lock);
1084 sched_annotate_sleep();
1087 * Check thread_group_leader() to exclude the traced sub-threads.
1089 if (state == EXIT_DEAD && thread_group_leader(p)) {
1090 struct signal_struct *sig = p->signal;
1091 struct signal_struct *psig = current->signal;
1092 unsigned long maxrss;
1093 u64 tgutime, tgstime;
1096 * The resource counters for the group leader are in its
1097 * own task_struct. Those for dead threads in the group
1098 * are in its signal_struct, as are those for the child
1099 * processes it has previously reaped. All these
1100 * accumulate in the parent's signal_struct c* fields.
1102 * We don't bother to take a lock here to protect these
1103 * p->signal fields because the whole thread group is dead
1104 * and nobody can change them.
1106 * psig->stats_lock also protects us from our sub-theads
1107 * which can reap other children at the same time. Until
1108 * we change k_getrusage()-like users to rely on this lock
1109 * we have to take ->siglock as well.
1111 * We use thread_group_cputime_adjusted() to get times for
1112 * the thread group, which consolidates times for all threads
1113 * in the group including the group leader.
1115 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1116 spin_lock_irq(¤t->sighand->siglock);
1117 write_seqlock(&psig->stats_lock);
1118 psig->cutime += tgutime + sig->cutime;
1119 psig->cstime += tgstime + sig->cstime;
1120 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1122 p->min_flt + sig->min_flt + sig->cmin_flt;
1124 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1126 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1128 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1130 task_io_get_inblock(p) +
1131 sig->inblock + sig->cinblock;
1133 task_io_get_oublock(p) +
1134 sig->oublock + sig->coublock;
1135 maxrss = max(sig->maxrss, sig->cmaxrss);
1136 if (psig->cmaxrss < maxrss)
1137 psig->cmaxrss = maxrss;
1138 task_io_accounting_add(&psig->ioac, &p->ioac);
1139 task_io_accounting_add(&psig->ioac, &sig->ioac);
1140 write_sequnlock(&psig->stats_lock);
1141 spin_unlock_irq(¤t->sighand->siglock);
1145 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1146 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1147 ? p->signal->group_exit_code : p->exit_code;
1148 wo->wo_stat = status;
1150 if (state == EXIT_TRACE) {
1151 write_lock_irq(&tasklist_lock);
1152 /* We dropped tasklist, ptracer could die and untrace */
1155 /* If parent wants a zombie, don't release it now */
1156 state = EXIT_ZOMBIE;
1157 if (do_notify_parent(p, p->exit_signal))
1159 p->exit_state = state;
1160 write_unlock_irq(&tasklist_lock);
1162 if (state == EXIT_DEAD)
1166 infop = wo->wo_info;
1168 if ((status & 0x7f) == 0) {
1169 infop->cause = CLD_EXITED;
1170 infop->status = status >> 8;
1172 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1173 infop->status = status & 0x7f;
1182 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1185 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1186 return &p->exit_code;
1188 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1189 return &p->signal->group_exit_code;
1195 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1197 * @ptrace: is the wait for ptrace
1198 * @p: task to wait for
1200 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1203 * read_lock(&tasklist_lock), which is released if return value is
1204 * non-zero. Also, grabs and releases @p->sighand->siglock.
1207 * 0 if wait condition didn't exist and search for other wait conditions
1208 * should continue. Non-zero return, -errno on failure and @p's pid on
1209 * success, implies that tasklist_lock is released and wait condition
1210 * search should terminate.
1212 static int wait_task_stopped(struct wait_opts *wo,
1213 int ptrace, struct task_struct *p)
1215 struct waitid_info *infop;
1216 int exit_code, *p_code, why;
1217 uid_t uid = 0; /* unneeded, required by compiler */
1221 * Traditionally we see ptrace'd stopped tasks regardless of options.
1223 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1226 if (!task_stopped_code(p, ptrace))
1230 spin_lock_irq(&p->sighand->siglock);
1232 p_code = task_stopped_code(p, ptrace);
1233 if (unlikely(!p_code))
1236 exit_code = *p_code;
1240 if (!unlikely(wo->wo_flags & WNOWAIT))
1243 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1245 spin_unlock_irq(&p->sighand->siglock);
1250 * Now we are pretty sure this task is interesting.
1251 * Make sure it doesn't get reaped out from under us while we
1252 * give up the lock and then examine it below. We don't want to
1253 * keep holding onto the tasklist_lock while we call getrusage and
1254 * possibly take page faults for user memory.
1257 pid = task_pid_vnr(p);
1258 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1259 read_unlock(&tasklist_lock);
1260 sched_annotate_sleep();
1262 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1265 if (likely(!(wo->wo_flags & WNOWAIT)))
1266 wo->wo_stat = (exit_code << 8) | 0x7f;
1268 infop = wo->wo_info;
1271 infop->status = exit_code;
1279 * Handle do_wait work for one task in a live, non-stopped state.
1280 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1281 * the lock and this task is uninteresting. If we return nonzero, we have
1282 * released the lock and the system call should return.
1284 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1286 struct waitid_info *infop;
1290 if (!unlikely(wo->wo_flags & WCONTINUED))
1293 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1296 spin_lock_irq(&p->sighand->siglock);
1297 /* Re-check with the lock held. */
1298 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1299 spin_unlock_irq(&p->sighand->siglock);
1302 if (!unlikely(wo->wo_flags & WNOWAIT))
1303 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1304 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1305 spin_unlock_irq(&p->sighand->siglock);
1307 pid = task_pid_vnr(p);
1309 read_unlock(&tasklist_lock);
1310 sched_annotate_sleep();
1312 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1315 infop = wo->wo_info;
1317 wo->wo_stat = 0xffff;
1319 infop->cause = CLD_CONTINUED;
1322 infop->status = SIGCONT;
1328 * Consider @p for a wait by @parent.
1330 * -ECHILD should be in ->notask_error before the first call.
1331 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1332 * Returns zero if the search for a child should continue;
1333 * then ->notask_error is 0 if @p is an eligible child,
1336 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1337 struct task_struct *p)
1340 * We can race with wait_task_zombie() from another thread.
1341 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1342 * can't confuse the checks below.
1344 int exit_state = ACCESS_ONCE(p->exit_state);
1347 if (unlikely(exit_state == EXIT_DEAD))
1350 ret = eligible_child(wo, ptrace, p);
1354 if (unlikely(exit_state == EXIT_TRACE)) {
1356 * ptrace == 0 means we are the natural parent. In this case
1357 * we should clear notask_error, debugger will notify us.
1359 if (likely(!ptrace))
1360 wo->notask_error = 0;
1364 if (likely(!ptrace) && unlikely(p->ptrace)) {
1366 * If it is traced by its real parent's group, just pretend
1367 * the caller is ptrace_do_wait() and reap this child if it
1370 * This also hides group stop state from real parent; otherwise
1371 * a single stop can be reported twice as group and ptrace stop.
1372 * If a ptracer wants to distinguish these two events for its
1373 * own children it should create a separate process which takes
1374 * the role of real parent.
1376 if (!ptrace_reparented(p))
1381 if (exit_state == EXIT_ZOMBIE) {
1382 /* we don't reap group leaders with subthreads */
1383 if (!delay_group_leader(p)) {
1385 * A zombie ptracee is only visible to its ptracer.
1386 * Notification and reaping will be cascaded to the
1387 * real parent when the ptracer detaches.
1389 if (unlikely(ptrace) || likely(!p->ptrace))
1390 return wait_task_zombie(wo, p);
1394 * Allow access to stopped/continued state via zombie by
1395 * falling through. Clearing of notask_error is complex.
1399 * If WEXITED is set, notask_error should naturally be
1400 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1401 * so, if there are live subthreads, there are events to
1402 * wait for. If all subthreads are dead, it's still safe
1403 * to clear - this function will be called again in finite
1404 * amount time once all the subthreads are released and
1405 * will then return without clearing.
1409 * Stopped state is per-task and thus can't change once the
1410 * target task dies. Only continued and exited can happen.
1411 * Clear notask_error if WCONTINUED | WEXITED.
1413 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1414 wo->notask_error = 0;
1417 * @p is alive and it's gonna stop, continue or exit, so
1418 * there always is something to wait for.
1420 wo->notask_error = 0;
1424 * Wait for stopped. Depending on @ptrace, different stopped state
1425 * is used and the two don't interact with each other.
1427 ret = wait_task_stopped(wo, ptrace, p);
1432 * Wait for continued. There's only one continued state and the
1433 * ptracer can consume it which can confuse the real parent. Don't
1434 * use WCONTINUED from ptracer. You don't need or want it.
1436 return wait_task_continued(wo, p);
1440 * Do the work of do_wait() for one thread in the group, @tsk.
1442 * -ECHILD should be in ->notask_error before the first call.
1443 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1444 * Returns zero if the search for a child should continue; then
1445 * ->notask_error is 0 if there were any eligible children,
1448 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1450 struct task_struct *p;
1452 list_for_each_entry(p, &tsk->children, sibling) {
1453 int ret = wait_consider_task(wo, 0, p);
1462 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1464 struct task_struct *p;
1466 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1467 int ret = wait_consider_task(wo, 1, p);
1476 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1477 int sync, void *key)
1479 struct wait_opts *wo = container_of(wait, struct wait_opts,
1481 struct task_struct *p = key;
1483 if (!eligible_pid(wo, p))
1486 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1489 return default_wake_function(wait, mode, sync, key);
1492 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1494 __wake_up_sync_key(&parent->signal->wait_chldexit,
1495 TASK_INTERRUPTIBLE, 1, p);
1498 static long do_wait(struct wait_opts *wo)
1500 struct task_struct *tsk;
1503 trace_sched_process_wait(wo->wo_pid);
1505 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1506 wo->child_wait.private = current;
1507 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1510 * If there is nothing that can match our criteria, just get out.
1511 * We will clear ->notask_error to zero if we see any child that
1512 * might later match our criteria, even if we are not able to reap
1515 wo->notask_error = -ECHILD;
1516 if ((wo->wo_type < PIDTYPE_MAX) &&
1517 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1520 set_current_state(TASK_INTERRUPTIBLE);
1521 read_lock(&tasklist_lock);
1524 retval = do_wait_thread(wo, tsk);
1528 retval = ptrace_do_wait(wo, tsk);
1532 if (wo->wo_flags & __WNOTHREAD)
1534 } while_each_thread(current, tsk);
1535 read_unlock(&tasklist_lock);
1538 retval = wo->notask_error;
1539 if (!retval && !(wo->wo_flags & WNOHANG)) {
1540 retval = -ERESTARTSYS;
1541 if (!signal_pending(current)) {
1547 __set_current_state(TASK_RUNNING);
1548 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1552 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1553 int options, struct rusage *ru)
1555 struct wait_opts wo;
1556 struct pid *pid = NULL;
1560 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1561 __WNOTHREAD|__WCLONE|__WALL))
1563 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1576 type = PIDTYPE_PGID;
1584 if (type < PIDTYPE_MAX)
1585 pid = find_get_pid(upid);
1589 wo.wo_flags = options;
1601 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1602 infop, int, options, struct rusage __user *, ru)
1605 struct waitid_info info = {.status = 0};
1606 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1609 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1615 user_access_begin();
1616 unsafe_put_user(err ? 0 : SIGCHLD, &infop->si_signo, Efault);
1617 unsafe_put_user(0, &infop->si_errno, Efault);
1618 unsafe_put_user((short)info.cause, &infop->si_code, Efault);
1619 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1620 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1621 unsafe_put_user(info.status, &infop->si_status, Efault);
1629 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1632 struct wait_opts wo;
1633 struct pid *pid = NULL;
1637 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1638 __WNOTHREAD|__WCLONE|__WALL))
1643 else if (upid < 0) {
1644 type = PIDTYPE_PGID;
1645 pid = find_get_pid(-upid);
1646 } else if (upid == 0) {
1647 type = PIDTYPE_PGID;
1648 pid = get_task_pid(current, PIDTYPE_PGID);
1649 } else /* upid > 0 */ {
1651 pid = find_get_pid(upid);
1656 wo.wo_flags = options | WEXITED;
1662 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1668 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1669 int, options, struct rusage __user *, ru)
1672 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1675 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1681 #ifdef __ARCH_WANT_SYS_WAITPID
1684 * sys_waitpid() remains for compatibility. waitpid() should be
1685 * implemented by calling sys_wait4() from libc.a.
1687 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1689 return sys_wait4(pid, stat_addr, options, NULL);
1694 #ifdef CONFIG_COMPAT
1695 COMPAT_SYSCALL_DEFINE4(wait4,
1697 compat_uint_t __user *, stat_addr,
1699 struct compat_rusage __user *, ru)
1702 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1704 if (ru && put_compat_rusage(&r, ru))
1710 COMPAT_SYSCALL_DEFINE5(waitid,
1711 int, which, compat_pid_t, pid,
1712 struct compat_siginfo __user *, infop, int, options,
1713 struct compat_rusage __user *, uru)
1716 struct waitid_info info = {.status = 0};
1717 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1720 /* kernel_waitid() overwrites everything in ru */
1721 if (COMPAT_USE_64BIT_TIME)
1722 err = copy_to_user(uru, &ru, sizeof(ru));
1724 err = put_compat_rusage(&ru, uru);
1732 user_access_begin();
1733 unsafe_put_user(err ? 0 : SIGCHLD, &infop->si_signo, Efault);
1734 unsafe_put_user(0, &infop->si_errno, Efault);
1735 unsafe_put_user((short)info.cause, &infop->si_code, Efault);
1736 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1737 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1738 unsafe_put_user(info.status, &infop->si_status, Efault);