1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 #include <linux/slab.h>
10 #include <linux/sched/autogroup.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/task.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/sched/cputime.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/capability.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/tty.h>
22 #include <linux/iocontext.h>
23 #include <linux/key.h>
24 #include <linux/cpu.h>
25 #include <linux/acct.h>
26 #include <linux/tsacct_kern.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/freezer.h>
30 #include <linux/binfmts.h>
31 #include <linux/nsproxy.h>
32 #include <linux/pid_namespace.h>
33 #include <linux/ptrace.h>
34 #include <linux/profile.h>
35 #include <linux/mount.h>
36 #include <linux/proc_fs.h>
37 #include <linux/kthread.h>
38 #include <linux/mempolicy.h>
39 #include <linux/taskstats_kern.h>
40 #include <linux/delayacct.h>
41 #include <linux/cgroup.h>
42 #include <linux/syscalls.h>
43 #include <linux/signal.h>
44 #include <linux/posix-timers.h>
45 #include <linux/cn_proc.h>
46 #include <linux/mutex.h>
47 #include <linux/futex.h>
48 #include <linux/pipe_fs_i.h>
49 #include <linux/audit.h> /* for audit_free() */
50 #include <linux/resource.h>
51 #include <linux/task_io_accounting_ops.h>
52 #include <linux/blkdev.h>
53 #include <linux/task_work.h>
54 #include <linux/fs_struct.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/kmsan.h>
64 #include <linux/random.h>
65 #include <linux/rcuwait.h>
66 #include <linux/compat.h>
67 #include <linux/io_uring.h>
68 #include <linux/kprobes.h>
69 #include <linux/rethook.h>
71 #include <linux/uaccess.h>
72 #include <asm/unistd.h>
73 #include <asm/mmu_context.h>
75 static void __unhash_process(struct task_struct *p, bool group_dead)
78 detach_pid(p, PIDTYPE_PID);
80 detach_pid(p, PIDTYPE_TGID);
81 detach_pid(p, PIDTYPE_PGID);
82 detach_pid(p, PIDTYPE_SID);
84 list_del_rcu(&p->tasks);
85 list_del_init(&p->sibling);
86 __this_cpu_dec(process_counts);
88 list_del_rcu(&p->thread_group);
89 list_del_rcu(&p->thread_node);
93 * This function expects the tasklist_lock write-locked.
95 static void __exit_signal(struct task_struct *tsk)
97 struct signal_struct *sig = tsk->signal;
98 bool group_dead = thread_group_leader(tsk);
99 struct sighand_struct *sighand;
100 struct tty_struct *tty;
103 sighand = rcu_dereference_check(tsk->sighand,
104 lockdep_tasklist_lock_is_held());
105 spin_lock(&sighand->siglock);
107 #ifdef CONFIG_POSIX_TIMERS
108 posix_cpu_timers_exit(tsk);
110 posix_cpu_timers_exit_group(tsk);
118 * If there is any task waiting for the group exit
121 if (sig->notify_count > 0 && !--sig->notify_count)
122 wake_up_process(sig->group_exec_task);
124 if (tsk == sig->curr_target)
125 sig->curr_target = next_thread(tsk);
128 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
129 sizeof(unsigned long long));
132 * Accumulate here the counters for all threads as they die. We could
133 * skip the group leader because it is the last user of signal_struct,
134 * but we want to avoid the race with thread_group_cputime() which can
135 * see the empty ->thread_head list.
137 task_cputime(tsk, &utime, &stime);
138 write_seqlock(&sig->stats_lock);
141 sig->gtime += task_gtime(tsk);
142 sig->min_flt += tsk->min_flt;
143 sig->maj_flt += tsk->maj_flt;
144 sig->nvcsw += tsk->nvcsw;
145 sig->nivcsw += tsk->nivcsw;
146 sig->inblock += task_io_get_inblock(tsk);
147 sig->oublock += task_io_get_oublock(tsk);
148 task_io_accounting_add(&sig->ioac, &tsk->ioac);
149 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
151 __unhash_process(tsk, group_dead);
152 write_sequnlock(&sig->stats_lock);
155 * Do this under ->siglock, we can race with another thread
156 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
158 flush_sigqueue(&tsk->pending);
160 spin_unlock(&sighand->siglock);
162 __cleanup_sighand(sighand);
163 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
165 flush_sigqueue(&sig->shared_pending);
170 static void delayed_put_task_struct(struct rcu_head *rhp)
172 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
174 kprobe_flush_task(tsk);
175 rethook_flush_task(tsk);
176 perf_event_delayed_put(tsk);
177 trace_sched_process_free(tsk);
178 put_task_struct(tsk);
181 void put_task_struct_rcu_user(struct task_struct *task)
183 if (refcount_dec_and_test(&task->rcu_users))
184 call_rcu(&task->rcu, delayed_put_task_struct);
187 void release_task(struct task_struct *p)
189 struct task_struct *leader;
190 struct pid *thread_pid;
193 /* don't need to get the RCU readlock here - the process is dead and
194 * can't be modifying its own credentials. But shut RCU-lockdep up */
196 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
201 write_lock_irq(&tasklist_lock);
202 ptrace_release_task(p);
203 thread_pid = get_pid(p->thread_pid);
207 * If we are the last non-leader member of the thread
208 * group, and the leader is zombie, then notify the
209 * group leader's parent process. (if it wants notification.)
212 leader = p->group_leader;
213 if (leader != p && thread_group_empty(leader)
214 && leader->exit_state == EXIT_ZOMBIE) {
216 * If we were the last child thread and the leader has
217 * exited already, and the leader's parent ignores SIGCHLD,
218 * then we are the one who should release the leader.
220 zap_leader = do_notify_parent(leader, leader->exit_signal);
222 leader->exit_state = EXIT_DEAD;
225 write_unlock_irq(&tasklist_lock);
226 seccomp_filter_release(p);
227 proc_flush_pid(thread_pid);
230 put_task_struct_rcu_user(p);
233 if (unlikely(zap_leader))
237 int rcuwait_wake_up(struct rcuwait *w)
240 struct task_struct *task;
245 * Order condition vs @task, such that everything prior to the load
246 * of @task is visible. This is the condition as to why the user called
247 * rcuwait_wake() in the first place. Pairs with set_current_state()
248 * barrier (A) in rcuwait_wait_event().
251 * [S] tsk = current [S] cond = true
257 task = rcu_dereference(w->task);
259 ret = wake_up_process(task);
264 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
267 * Determine if a process group is "orphaned", according to the POSIX
268 * definition in 2.2.2.52. Orphaned process groups are not to be affected
269 * by terminal-generated stop signals. Newly orphaned process groups are
270 * to receive a SIGHUP and a SIGCONT.
272 * "I ask you, have you ever known what it is to be an orphan?"
274 static int will_become_orphaned_pgrp(struct pid *pgrp,
275 struct task_struct *ignored_task)
277 struct task_struct *p;
279 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
280 if ((p == ignored_task) ||
281 (p->exit_state && thread_group_empty(p)) ||
282 is_global_init(p->real_parent))
285 if (task_pgrp(p->real_parent) != pgrp &&
286 task_session(p->real_parent) == task_session(p))
288 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
293 int is_current_pgrp_orphaned(void)
297 read_lock(&tasklist_lock);
298 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
299 read_unlock(&tasklist_lock);
304 static bool has_stopped_jobs(struct pid *pgrp)
306 struct task_struct *p;
308 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
309 if (p->signal->flags & SIGNAL_STOP_STOPPED)
311 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
317 * Check to see if any process groups have become orphaned as
318 * a result of our exiting, and if they have any stopped jobs,
319 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
322 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
324 struct pid *pgrp = task_pgrp(tsk);
325 struct task_struct *ignored_task = tsk;
328 /* exit: our father is in a different pgrp than
329 * we are and we were the only connection outside.
331 parent = tsk->real_parent;
333 /* reparent: our child is in a different pgrp than
334 * we are, and it was the only connection outside.
338 if (task_pgrp(parent) != pgrp &&
339 task_session(parent) == task_session(tsk) &&
340 will_become_orphaned_pgrp(pgrp, ignored_task) &&
341 has_stopped_jobs(pgrp)) {
342 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
343 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
347 static void coredump_task_exit(struct task_struct *tsk)
349 struct core_state *core_state;
352 * Serialize with any possible pending coredump.
353 * We must hold siglock around checking core_state
354 * and setting PF_POSTCOREDUMP. The core-inducing thread
355 * will increment ->nr_threads for each thread in the
356 * group without PF_POSTCOREDUMP set.
358 spin_lock_irq(&tsk->sighand->siglock);
359 tsk->flags |= PF_POSTCOREDUMP;
360 core_state = tsk->signal->core_state;
361 spin_unlock_irq(&tsk->sighand->siglock);
363 struct core_thread self;
366 if (self.task->flags & PF_SIGNALED)
367 self.next = xchg(&core_state->dumper.next, &self);
371 * Implies mb(), the result of xchg() must be visible
372 * to core_state->dumper.
374 if (atomic_dec_and_test(&core_state->nr_threads))
375 complete(&core_state->startup);
378 set_current_state(TASK_UNINTERRUPTIBLE|TASK_FREEZABLE);
379 if (!self.task) /* see coredump_finish() */
383 __set_current_state(TASK_RUNNING);
389 * A task is exiting. If it owned this mm, find a new owner for the mm.
391 void mm_update_next_owner(struct mm_struct *mm)
393 struct task_struct *c, *g, *p = current;
397 * If the exiting or execing task is not the owner, it's
398 * someone else's problem.
403 * The current owner is exiting/execing and there are no other
404 * candidates. Do not leave the mm pointing to a possibly
405 * freed task structure.
407 if (atomic_read(&mm->mm_users) <= 1) {
408 WRITE_ONCE(mm->owner, NULL);
412 read_lock(&tasklist_lock);
414 * Search in the children
416 list_for_each_entry(c, &p->children, sibling) {
418 goto assign_new_owner;
422 * Search in the siblings
424 list_for_each_entry(c, &p->real_parent->children, sibling) {
426 goto assign_new_owner;
430 * Search through everything else, we should not get here often.
432 for_each_process(g) {
433 if (g->flags & PF_KTHREAD)
435 for_each_thread(g, c) {
437 goto assign_new_owner;
442 read_unlock(&tasklist_lock);
444 * We found no owner yet mm_users > 1: this implies that we are
445 * most likely racing with swapoff (try_to_unuse()) or /proc or
446 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
448 WRITE_ONCE(mm->owner, NULL);
455 * The task_lock protects c->mm from changing.
456 * We always want mm->owner->mm == mm
460 * Delay read_unlock() till we have the task_lock()
461 * to ensure that c does not slip away underneath us
463 read_unlock(&tasklist_lock);
469 WRITE_ONCE(mm->owner, c);
470 lru_gen_migrate_mm(mm);
474 #endif /* CONFIG_MEMCG */
477 * Turn us into a lazy TLB process if we
480 static void exit_mm(void)
482 struct mm_struct *mm = current->mm;
484 exit_mm_release(current, mm);
490 BUG_ON(mm != current->active_mm);
491 /* more a memory barrier than a real lock */
494 * When a thread stops operating on an address space, the loop
495 * in membarrier_private_expedited() may not observe that
496 * tsk->mm, and the loop in membarrier_global_expedited() may
497 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
498 * rq->membarrier_state, so those would not issue an IPI.
499 * Membarrier requires a memory barrier after accessing
500 * user-space memory, before clearing tsk->mm or the
501 * rq->membarrier_state.
503 smp_mb__after_spinlock();
506 membarrier_update_current_mm(NULL);
507 enter_lazy_tlb(mm, current);
509 task_unlock(current);
510 mmap_read_unlock(mm);
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);
548 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
549 list_del_init(&p->ptrace_entry);
553 zap_pid_ns_processes(pid_ns);
554 write_lock_irq(&tasklist_lock);
560 * When we die, we re-parent all our children, and try to:
561 * 1. give them to another thread in our thread group, if such a member exists
562 * 2. give it to the first ancestor process which prctl'd itself as a
563 * child_subreaper for its children (like a service manager)
564 * 3. give it to the init process (PID 1) in our pid namespace
566 static struct task_struct *find_new_reaper(struct task_struct *father,
567 struct task_struct *child_reaper)
569 struct task_struct *thread, *reaper;
571 thread = find_alive_thread(father);
575 if (father->signal->has_child_subreaper) {
576 unsigned int ns_level = task_pid(father)->level;
578 * Find the first ->is_child_subreaper ancestor in our pid_ns.
579 * We can't check reaper != child_reaper to ensure we do not
580 * cross the namespaces, the exiting parent could be injected
581 * by setns() + fork().
582 * We check pid->level, this is slightly more efficient than
583 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
585 for (reaper = father->real_parent;
586 task_pid(reaper)->level == ns_level;
587 reaper = reaper->real_parent) {
588 if (reaper == &init_task)
590 if (!reaper->signal->is_child_subreaper)
592 thread = find_alive_thread(reaper);
602 * Any that need to be release_task'd are put on the @dead list.
604 static void reparent_leader(struct task_struct *father, struct task_struct *p,
605 struct list_head *dead)
607 if (unlikely(p->exit_state == EXIT_DEAD))
610 /* We don't want people slaying init. */
611 p->exit_signal = SIGCHLD;
613 /* If it has exited notify the new parent about this child's death. */
615 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
616 if (do_notify_parent(p, p->exit_signal)) {
617 p->exit_state = EXIT_DEAD;
618 list_add(&p->ptrace_entry, dead);
622 kill_orphaned_pgrp(p, father);
626 * This does two things:
628 * A. Make init inherit all the child processes
629 * B. Check to see if any process groups have become orphaned
630 * as a result of our exiting, and if they have any stopped
631 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
633 static void forget_original_parent(struct task_struct *father,
634 struct list_head *dead)
636 struct task_struct *p, *t, *reaper;
638 if (unlikely(!list_empty(&father->ptraced)))
639 exit_ptrace(father, dead);
641 /* Can drop and reacquire tasklist_lock */
642 reaper = find_child_reaper(father, dead);
643 if (list_empty(&father->children))
646 reaper = find_new_reaper(father, reaper);
647 list_for_each_entry(p, &father->children, sibling) {
648 for_each_thread(p, t) {
649 RCU_INIT_POINTER(t->real_parent, reaper);
650 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
651 if (likely(!t->ptrace))
652 t->parent = t->real_parent;
653 if (t->pdeath_signal)
654 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 tsk->exit_state = EXIT_ZOMBIE;
685 if (unlikely(tsk->ptrace)) {
686 int sig = thread_group_leader(tsk) &&
687 thread_group_empty(tsk) &&
688 !ptrace_reparented(tsk) ?
689 tsk->exit_signal : SIGCHLD;
690 autoreap = do_notify_parent(tsk, sig);
691 } else if (thread_group_leader(tsk)) {
692 autoreap = thread_group_empty(tsk) &&
693 do_notify_parent(tsk, tsk->exit_signal);
699 tsk->exit_state = EXIT_DEAD;
700 list_add(&tsk->ptrace_entry, &dead);
703 /* mt-exec, de_thread() is waiting for group leader */
704 if (unlikely(tsk->signal->notify_count < 0))
705 wake_up_process(tsk->signal->group_exec_task);
706 write_unlock_irq(&tasklist_lock);
708 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
709 list_del_init(&p->ptrace_entry);
714 #ifdef CONFIG_DEBUG_STACK_USAGE
715 static void check_stack_usage(void)
717 static DEFINE_SPINLOCK(low_water_lock);
718 static int lowest_to_date = THREAD_SIZE;
721 free = stack_not_used(current);
723 if (free >= lowest_to_date)
726 spin_lock(&low_water_lock);
727 if (free < lowest_to_date) {
728 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
729 current->comm, task_pid_nr(current), free);
730 lowest_to_date = free;
732 spin_unlock(&low_water_lock);
735 static inline void check_stack_usage(void) {}
738 static void synchronize_group_exit(struct task_struct *tsk, long code)
740 struct sighand_struct *sighand = tsk->sighand;
741 struct signal_struct *signal = tsk->signal;
743 spin_lock_irq(&sighand->siglock);
744 signal->quick_threads--;
745 if ((signal->quick_threads == 0) &&
746 !(signal->flags & SIGNAL_GROUP_EXIT)) {
747 signal->flags = SIGNAL_GROUP_EXIT;
748 signal->group_exit_code = code;
749 signal->group_stop_count = 0;
751 spin_unlock_irq(&sighand->siglock);
754 void __noreturn do_exit(long code)
756 struct task_struct *tsk = current;
759 synchronize_group_exit(tsk, code);
764 kmsan_task_exit(tsk);
766 coredump_task_exit(tsk);
767 ptrace_event(PTRACE_EVENT_EXIT, code);
769 validate_creds_for_do_exit(tsk);
771 io_uring_files_cancel();
772 exit_signals(tsk); /* sets PF_EXITING */
774 /* sync mm's RSS info before statistics gathering */
776 sync_mm_rss(tsk->mm);
777 acct_update_integrals(tsk);
778 group_dead = atomic_dec_and_test(&tsk->signal->live);
781 * If the last thread of global init has exited, panic
782 * immediately to get a useable coredump.
784 if (unlikely(is_global_init(tsk)))
785 panic("Attempted to kill init! exitcode=0x%08x\n",
786 tsk->signal->group_exit_code ?: (int)code);
788 #ifdef CONFIG_POSIX_TIMERS
789 hrtimer_cancel(&tsk->signal->real_timer);
793 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
795 acct_collect(code, group_dead);
800 tsk->exit_code = code;
801 taskstats_exit(tsk, group_dead);
807 trace_sched_process_exit(tsk);
814 disassociate_ctty(1);
815 exit_task_namespaces(tsk);
820 * Flush inherited counters to the parent - before the parent
821 * gets woken up by child-exit notifications.
823 * because of cgroup mode, must be called before cgroup_exit()
825 perf_event_exit_task(tsk);
827 sched_autogroup_exit_task(tsk);
831 * FIXME: do that only when needed, using sched_exit tracepoint
833 flush_ptrace_hw_breakpoint(tsk);
835 exit_tasks_rcu_start();
836 exit_notify(tsk, group_dead);
837 proc_exit_connector(tsk);
838 mpol_put_task_policy(tsk);
840 if (unlikely(current->pi_state_cache))
841 kfree(current->pi_state_cache);
844 * Make sure we are holding no locks:
846 debug_check_no_locks_held();
849 exit_io_context(tsk);
851 if (tsk->splice_pipe)
852 free_pipe_info(tsk->splice_pipe);
854 if (tsk->task_frag.page)
855 put_page(tsk->task_frag.page);
857 validate_creds_for_do_exit(tsk);
858 exit_task_stack_account(tsk);
863 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
865 exit_tasks_rcu_finish();
867 lockdep_free_task(tsk);
871 void __noreturn make_task_dead(int signr)
874 * Take the task off the cpu after something catastrophic has
877 * We can get here from a kernel oops, sometimes with preemption off.
878 * Start by checking for critical errors.
879 * Then fix up important state like USER_DS and preemption.
880 * Then do everything else.
882 struct task_struct *tsk = current;
884 if (unlikely(in_interrupt()))
885 panic("Aiee, killing interrupt handler!");
886 if (unlikely(!tsk->pid))
887 panic("Attempted to kill the idle task!");
889 if (unlikely(in_atomic())) {
890 pr_info("note: %s[%d] exited with preempt_count %d\n",
891 current->comm, task_pid_nr(current),
893 preempt_count_set(PREEMPT_ENABLED);
897 * We're taking recursive faults here in make_task_dead. Safest is to just
898 * leave this task alone and wait for reboot.
900 if (unlikely(tsk->flags & PF_EXITING)) {
901 pr_alert("Fixing recursive fault but reboot is needed!\n");
902 futex_exit_recursive(tsk);
903 tsk->exit_state = EXIT_DEAD;
904 refcount_inc(&tsk->rcu_users);
911 SYSCALL_DEFINE1(exit, int, error_code)
913 do_exit((error_code&0xff)<<8);
917 * Take down every thread in the group. This is called by fatal signals
918 * as well as by sys_exit_group (below).
921 do_group_exit(int exit_code)
923 struct signal_struct *sig = current->signal;
925 if (sig->flags & SIGNAL_GROUP_EXIT)
926 exit_code = sig->group_exit_code;
927 else if (sig->group_exec_task)
930 struct sighand_struct *const sighand = current->sighand;
932 spin_lock_irq(&sighand->siglock);
933 if (sig->flags & SIGNAL_GROUP_EXIT)
934 /* Another thread got here before we took the lock. */
935 exit_code = sig->group_exit_code;
936 else if (sig->group_exec_task)
939 sig->group_exit_code = exit_code;
940 sig->flags = SIGNAL_GROUP_EXIT;
941 zap_other_threads(current);
943 spin_unlock_irq(&sighand->siglock);
951 * this kills every thread in the thread group. Note that any externally
952 * wait4()-ing process will get the correct exit code - even if this
953 * thread is not the thread group leader.
955 SYSCALL_DEFINE1(exit_group, int, error_code)
957 do_group_exit((error_code & 0xff) << 8);
970 enum pid_type wo_type;
974 struct waitid_info *wo_info;
976 struct rusage *wo_rusage;
978 wait_queue_entry_t child_wait;
982 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
984 return wo->wo_type == PIDTYPE_MAX ||
985 task_pid_type(p, wo->wo_type) == wo->wo_pid;
989 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
991 if (!eligible_pid(wo, p))
995 * Wait for all children (clone and not) if __WALL is set or
996 * if it is traced by us.
998 if (ptrace || (wo->wo_flags & __WALL))
1002 * Otherwise, wait for clone children *only* if __WCLONE is set;
1003 * otherwise, wait for non-clone children *only*.
1005 * Note: a "clone" child here is one that reports to its parent
1006 * using a signal other than SIGCHLD, or a non-leader thread which
1007 * we can only see if it is traced by us.
1009 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1016 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1017 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1018 * the lock and this task is uninteresting. If we return nonzero, we have
1019 * released the lock and the system call should return.
1021 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1024 pid_t pid = task_pid_vnr(p);
1025 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1026 struct waitid_info *infop;
1028 if (!likely(wo->wo_flags & WEXITED))
1031 if (unlikely(wo->wo_flags & WNOWAIT)) {
1032 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1033 ? p->signal->group_exit_code : p->exit_code;
1035 read_unlock(&tasklist_lock);
1036 sched_annotate_sleep();
1038 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1043 * Move the task's state to DEAD/TRACE, only one thread can do this.
1045 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1046 EXIT_TRACE : EXIT_DEAD;
1047 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1050 * We own this thread, nobody else can reap it.
1052 read_unlock(&tasklist_lock);
1053 sched_annotate_sleep();
1056 * Check thread_group_leader() to exclude the traced sub-threads.
1058 if (state == EXIT_DEAD && thread_group_leader(p)) {
1059 struct signal_struct *sig = p->signal;
1060 struct signal_struct *psig = current->signal;
1061 unsigned long maxrss;
1062 u64 tgutime, tgstime;
1065 * The resource counters for the group leader are in its
1066 * own task_struct. Those for dead threads in the group
1067 * are in its signal_struct, as are those for the child
1068 * processes it has previously reaped. All these
1069 * accumulate in the parent's signal_struct c* fields.
1071 * We don't bother to take a lock here to protect these
1072 * p->signal fields because the whole thread group is dead
1073 * and nobody can change them.
1075 * psig->stats_lock also protects us from our sub-threads
1076 * which can reap other children at the same time. Until
1077 * we change k_getrusage()-like users to rely on this lock
1078 * we have to take ->siglock as well.
1080 * We use thread_group_cputime_adjusted() to get times for
1081 * the thread group, which consolidates times for all threads
1082 * in the group including the group leader.
1084 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1085 spin_lock_irq(¤t->sighand->siglock);
1086 write_seqlock(&psig->stats_lock);
1087 psig->cutime += tgutime + sig->cutime;
1088 psig->cstime += tgstime + sig->cstime;
1089 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1091 p->min_flt + sig->min_flt + sig->cmin_flt;
1093 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1095 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1097 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1099 task_io_get_inblock(p) +
1100 sig->inblock + sig->cinblock;
1102 task_io_get_oublock(p) +
1103 sig->oublock + sig->coublock;
1104 maxrss = max(sig->maxrss, sig->cmaxrss);
1105 if (psig->cmaxrss < maxrss)
1106 psig->cmaxrss = maxrss;
1107 task_io_accounting_add(&psig->ioac, &p->ioac);
1108 task_io_accounting_add(&psig->ioac, &sig->ioac);
1109 write_sequnlock(&psig->stats_lock);
1110 spin_unlock_irq(¤t->sighand->siglock);
1114 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1115 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1116 ? p->signal->group_exit_code : p->exit_code;
1117 wo->wo_stat = status;
1119 if (state == EXIT_TRACE) {
1120 write_lock_irq(&tasklist_lock);
1121 /* We dropped tasklist, ptracer could die and untrace */
1124 /* If parent wants a zombie, don't release it now */
1125 state = EXIT_ZOMBIE;
1126 if (do_notify_parent(p, p->exit_signal))
1128 p->exit_state = state;
1129 write_unlock_irq(&tasklist_lock);
1131 if (state == EXIT_DEAD)
1135 infop = wo->wo_info;
1137 if ((status & 0x7f) == 0) {
1138 infop->cause = CLD_EXITED;
1139 infop->status = status >> 8;
1141 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1142 infop->status = status & 0x7f;
1151 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1154 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1155 return &p->exit_code;
1157 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1158 return &p->signal->group_exit_code;
1164 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1166 * @ptrace: is the wait for ptrace
1167 * @p: task to wait for
1169 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1172 * read_lock(&tasklist_lock), which is released if return value is
1173 * non-zero. Also, grabs and releases @p->sighand->siglock.
1176 * 0 if wait condition didn't exist and search for other wait conditions
1177 * should continue. Non-zero return, -errno on failure and @p's pid on
1178 * success, implies that tasklist_lock is released and wait condition
1179 * search should terminate.
1181 static int wait_task_stopped(struct wait_opts *wo,
1182 int ptrace, struct task_struct *p)
1184 struct waitid_info *infop;
1185 int exit_code, *p_code, why;
1186 uid_t uid = 0; /* unneeded, required by compiler */
1190 * Traditionally we see ptrace'd stopped tasks regardless of options.
1192 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1195 if (!task_stopped_code(p, ptrace))
1199 spin_lock_irq(&p->sighand->siglock);
1201 p_code = task_stopped_code(p, ptrace);
1202 if (unlikely(!p_code))
1205 exit_code = *p_code;
1209 if (!unlikely(wo->wo_flags & WNOWAIT))
1212 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1214 spin_unlock_irq(&p->sighand->siglock);
1219 * Now we are pretty sure this task is interesting.
1220 * Make sure it doesn't get reaped out from under us while we
1221 * give up the lock and then examine it below. We don't want to
1222 * keep holding onto the tasklist_lock while we call getrusage and
1223 * possibly take page faults for user memory.
1226 pid = task_pid_vnr(p);
1227 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1228 read_unlock(&tasklist_lock);
1229 sched_annotate_sleep();
1231 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1234 if (likely(!(wo->wo_flags & WNOWAIT)))
1235 wo->wo_stat = (exit_code << 8) | 0x7f;
1237 infop = wo->wo_info;
1240 infop->status = exit_code;
1248 * Handle do_wait work for one task in a live, non-stopped state.
1249 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1250 * the lock and this task is uninteresting. If we return nonzero, we have
1251 * released the lock and the system call should return.
1253 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1255 struct waitid_info *infop;
1259 if (!unlikely(wo->wo_flags & WCONTINUED))
1262 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1265 spin_lock_irq(&p->sighand->siglock);
1266 /* Re-check with the lock held. */
1267 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1268 spin_unlock_irq(&p->sighand->siglock);
1271 if (!unlikely(wo->wo_flags & WNOWAIT))
1272 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1273 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1274 spin_unlock_irq(&p->sighand->siglock);
1276 pid = task_pid_vnr(p);
1278 read_unlock(&tasklist_lock);
1279 sched_annotate_sleep();
1281 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1284 infop = wo->wo_info;
1286 wo->wo_stat = 0xffff;
1288 infop->cause = CLD_CONTINUED;
1291 infop->status = SIGCONT;
1297 * Consider @p for a wait by @parent.
1299 * -ECHILD should be in ->notask_error before the first call.
1300 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1301 * Returns zero if the search for a child should continue;
1302 * then ->notask_error is 0 if @p is an eligible child,
1305 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1306 struct task_struct *p)
1309 * We can race with wait_task_zombie() from another thread.
1310 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1311 * can't confuse the checks below.
1313 int exit_state = READ_ONCE(p->exit_state);
1316 if (unlikely(exit_state == EXIT_DEAD))
1319 ret = eligible_child(wo, ptrace, p);
1323 if (unlikely(exit_state == EXIT_TRACE)) {
1325 * ptrace == 0 means we are the natural parent. In this case
1326 * we should clear notask_error, debugger will notify us.
1328 if (likely(!ptrace))
1329 wo->notask_error = 0;
1333 if (likely(!ptrace) && unlikely(p->ptrace)) {
1335 * If it is traced by its real parent's group, just pretend
1336 * the caller is ptrace_do_wait() and reap this child if it
1339 * This also hides group stop state from real parent; otherwise
1340 * a single stop can be reported twice as group and ptrace stop.
1341 * If a ptracer wants to distinguish these two events for its
1342 * own children it should create a separate process which takes
1343 * the role of real parent.
1345 if (!ptrace_reparented(p))
1350 if (exit_state == EXIT_ZOMBIE) {
1351 /* we don't reap group leaders with subthreads */
1352 if (!delay_group_leader(p)) {
1354 * A zombie ptracee is only visible to its ptracer.
1355 * Notification and reaping will be cascaded to the
1356 * real parent when the ptracer detaches.
1358 if (unlikely(ptrace) || likely(!p->ptrace))
1359 return wait_task_zombie(wo, p);
1363 * Allow access to stopped/continued state via zombie by
1364 * falling through. Clearing of notask_error is complex.
1368 * If WEXITED is set, notask_error should naturally be
1369 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1370 * so, if there are live subthreads, there are events to
1371 * wait for. If all subthreads are dead, it's still safe
1372 * to clear - this function will be called again in finite
1373 * amount time once all the subthreads are released and
1374 * will then return without clearing.
1378 * Stopped state is per-task and thus can't change once the
1379 * target task dies. Only continued and exited can happen.
1380 * Clear notask_error if WCONTINUED | WEXITED.
1382 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1383 wo->notask_error = 0;
1386 * @p is alive and it's gonna stop, continue or exit, so
1387 * there always is something to wait for.
1389 wo->notask_error = 0;
1393 * Wait for stopped. Depending on @ptrace, different stopped state
1394 * is used and the two don't interact with each other.
1396 ret = wait_task_stopped(wo, ptrace, p);
1401 * Wait for continued. There's only one continued state and the
1402 * ptracer can consume it which can confuse the real parent. Don't
1403 * use WCONTINUED from ptracer. You don't need or want it.
1405 return wait_task_continued(wo, p);
1409 * Do the work of do_wait() for one thread in the group, @tsk.
1411 * -ECHILD should be in ->notask_error before the first call.
1412 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1413 * Returns zero if the search for a child should continue; then
1414 * ->notask_error is 0 if there were any eligible children,
1417 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1419 struct task_struct *p;
1421 list_for_each_entry(p, &tsk->children, sibling) {
1422 int ret = wait_consider_task(wo, 0, p);
1431 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1433 struct task_struct *p;
1435 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1436 int ret = wait_consider_task(wo, 1, p);
1445 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1446 int sync, void *key)
1448 struct wait_opts *wo = container_of(wait, struct wait_opts,
1450 struct task_struct *p = key;
1452 if (!eligible_pid(wo, p))
1455 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1458 return default_wake_function(wait, mode, sync, key);
1461 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1463 __wake_up_sync_key(&parent->signal->wait_chldexit,
1464 TASK_INTERRUPTIBLE, p);
1467 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1468 struct task_struct *target)
1470 struct task_struct *parent =
1471 !ptrace ? target->real_parent : target->parent;
1473 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1474 same_thread_group(current, parent));
1478 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1479 * and tracee lists to find the target task.
1481 static int do_wait_pid(struct wait_opts *wo)
1484 struct task_struct *target;
1488 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1489 if (target && is_effectively_child(wo, ptrace, target)) {
1490 retval = wait_consider_task(wo, ptrace, target);
1496 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1497 if (target && target->ptrace &&
1498 is_effectively_child(wo, ptrace, target)) {
1499 retval = wait_consider_task(wo, ptrace, target);
1507 static long do_wait(struct wait_opts *wo)
1511 trace_sched_process_wait(wo->wo_pid);
1513 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1514 wo->child_wait.private = current;
1515 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1518 * If there is nothing that can match our criteria, just get out.
1519 * We will clear ->notask_error to zero if we see any child that
1520 * might later match our criteria, even if we are not able to reap
1523 wo->notask_error = -ECHILD;
1524 if ((wo->wo_type < PIDTYPE_MAX) &&
1525 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1528 set_current_state(TASK_INTERRUPTIBLE);
1529 read_lock(&tasklist_lock);
1531 if (wo->wo_type == PIDTYPE_PID) {
1532 retval = do_wait_pid(wo);
1536 struct task_struct *tsk = current;
1539 retval = do_wait_thread(wo, tsk);
1543 retval = ptrace_do_wait(wo, tsk);
1547 if (wo->wo_flags & __WNOTHREAD)
1549 } while_each_thread(current, tsk);
1551 read_unlock(&tasklist_lock);
1554 retval = wo->notask_error;
1555 if (!retval && !(wo->wo_flags & WNOHANG)) {
1556 retval = -ERESTARTSYS;
1557 if (!signal_pending(current)) {
1563 __set_current_state(TASK_RUNNING);
1564 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1568 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1569 int options, struct rusage *ru)
1571 struct wait_opts wo;
1572 struct pid *pid = NULL;
1575 unsigned int f_flags = 0;
1577 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1578 __WNOTHREAD|__WCLONE|__WALL))
1580 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1592 pid = find_get_pid(upid);
1595 type = PIDTYPE_PGID;
1600 pid = find_get_pid(upid);
1602 pid = get_task_pid(current, PIDTYPE_PGID);
1609 pid = pidfd_get_pid(upid, &f_flags);
1611 return PTR_ERR(pid);
1620 wo.wo_flags = options;
1623 if (f_flags & O_NONBLOCK)
1624 wo.wo_flags |= WNOHANG;
1627 if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1634 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1635 infop, int, options, struct rusage __user *, ru)
1638 struct waitid_info info = {.status = 0};
1639 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1645 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1651 if (!user_write_access_begin(infop, sizeof(*infop)))
1654 unsafe_put_user(signo, &infop->si_signo, Efault);
1655 unsafe_put_user(0, &infop->si_errno, Efault);
1656 unsafe_put_user(info.cause, &infop->si_code, Efault);
1657 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1658 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1659 unsafe_put_user(info.status, &infop->si_status, Efault);
1660 user_write_access_end();
1663 user_write_access_end();
1667 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1670 struct wait_opts wo;
1671 struct pid *pid = NULL;
1675 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1676 __WNOTHREAD|__WCLONE|__WALL))
1679 /* -INT_MIN is not defined */
1680 if (upid == INT_MIN)
1685 else if (upid < 0) {
1686 type = PIDTYPE_PGID;
1687 pid = find_get_pid(-upid);
1688 } else if (upid == 0) {
1689 type = PIDTYPE_PGID;
1690 pid = get_task_pid(current, PIDTYPE_PGID);
1691 } else /* upid > 0 */ {
1693 pid = find_get_pid(upid);
1698 wo.wo_flags = options | WEXITED;
1704 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1710 int kernel_wait(pid_t pid, int *stat)
1712 struct wait_opts wo = {
1713 .wo_type = PIDTYPE_PID,
1714 .wo_pid = find_get_pid(pid),
1715 .wo_flags = WEXITED,
1720 if (ret > 0 && wo.wo_stat)
1726 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1727 int, options, struct rusage __user *, ru)
1730 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1733 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1739 #ifdef __ARCH_WANT_SYS_WAITPID
1742 * sys_waitpid() remains for compatibility. waitpid() should be
1743 * implemented by calling sys_wait4() from libc.a.
1745 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1747 return kernel_wait4(pid, stat_addr, options, NULL);
1752 #ifdef CONFIG_COMPAT
1753 COMPAT_SYSCALL_DEFINE4(wait4,
1755 compat_uint_t __user *, stat_addr,
1757 struct compat_rusage __user *, ru)
1760 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1762 if (ru && put_compat_rusage(&r, ru))
1768 COMPAT_SYSCALL_DEFINE5(waitid,
1769 int, which, compat_pid_t, pid,
1770 struct compat_siginfo __user *, infop, int, options,
1771 struct compat_rusage __user *, uru)
1774 struct waitid_info info = {.status = 0};
1775 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1781 /* kernel_waitid() overwrites everything in ru */
1782 if (COMPAT_USE_64BIT_TIME)
1783 err = copy_to_user(uru, &ru, sizeof(ru));
1785 err = put_compat_rusage(&ru, uru);
1794 if (!user_write_access_begin(infop, sizeof(*infop)))
1797 unsafe_put_user(signo, &infop->si_signo, Efault);
1798 unsafe_put_user(0, &infop->si_errno, Efault);
1799 unsafe_put_user(info.cause, &infop->si_code, Efault);
1800 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1801 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1802 unsafe_put_user(info.status, &infop->si_status, Efault);
1803 user_write_access_end();
1806 user_write_access_end();
1812 * thread_group_exited - check that a thread group has exited
1813 * @pid: tgid of thread group to be checked.
1815 * Test if the thread group represented by tgid has exited (all
1816 * threads are zombies, dead or completely gone).
1818 * Return: true if the thread group has exited. false otherwise.
1820 bool thread_group_exited(struct pid *pid)
1822 struct task_struct *task;
1826 task = pid_task(pid, PIDTYPE_PID);
1828 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1833 EXPORT_SYMBOL(thread_group_exited);
1835 __weak void abort(void)
1839 /* if that doesn't kill us, halt */
1840 panic("Oops failed to kill thread");
1842 EXPORT_SYMBOL(abort);