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_TGID);
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);
199 write_lock_irq(&tasklist_lock);
200 ptrace_release_task(p);
204 * If we are the last non-leader member of the thread
205 * group, and the leader is zombie, then notify the
206 * group leader's parent process. (if it wants notification.)
209 leader = p->group_leader;
210 if (leader != p && thread_group_empty(leader)
211 && leader->exit_state == EXIT_ZOMBIE) {
213 * If we were the last child thread and the leader has
214 * exited already, and the leader's parent ignores SIGCHLD,
215 * then we are the one who should release the leader.
217 zap_leader = do_notify_parent(leader, leader->exit_signal);
219 leader->exit_state = EXIT_DEAD;
222 write_unlock_irq(&tasklist_lock);
224 call_rcu(&p->rcu, delayed_put_task_struct);
227 if (unlikely(zap_leader))
232 * Note that if this function returns a valid task_struct pointer (!NULL)
233 * task->usage must remain >0 for the duration of the RCU critical section.
235 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
237 struct sighand_struct *sighand;
238 struct task_struct *task;
241 * We need to verify that release_task() was not called and thus
242 * delayed_put_task_struct() can't run and drop the last reference
243 * before rcu_read_unlock(). We check task->sighand != NULL,
244 * but we can read the already freed and reused memory.
247 task = rcu_dereference(*ptask);
251 probe_kernel_address(&task->sighand, sighand);
254 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
255 * was already freed we can not miss the preceding update of this
259 if (unlikely(task != READ_ONCE(*ptask)))
263 * We've re-checked that "task == *ptask", now we have two different
266 * 1. This is actually the same task/task_struct. In this case
267 * sighand != NULL tells us it is still alive.
269 * 2. This is another task which got the same memory for task_struct.
270 * We can't know this of course, and we can not trust
273 * In this case we actually return a random value, but this is
276 * If we return NULL - we can pretend that we actually noticed that
277 * *ptask was updated when the previous task has exited. Or pretend
278 * that probe_slab_address(&sighand) reads NULL.
280 * If we return the new task (because sighand is not NULL for any
281 * reason) - this is fine too. This (new) task can't go away before
284 * And note: We could even eliminate the false positive if re-read
285 * task->sighand once again to avoid the falsely NULL. But this case
286 * is very unlikely so we don't care.
294 void rcuwait_wake_up(struct rcuwait *w)
296 struct task_struct *task;
301 * Order condition vs @task, such that everything prior to the load
302 * of @task is visible. This is the condition as to why the user called
303 * rcuwait_trywake() in the first place. Pairs with set_current_state()
304 * barrier (A) in rcuwait_wait_event().
307 * [S] tsk = current [S] cond = true
314 * Avoid using task_rcu_dereference() magic as long as we are careful,
315 * see comment in rcuwait_wait_event() regarding ->exit_state.
317 task = rcu_dereference(w->task);
319 wake_up_process(task);
324 * Determine if a process group is "orphaned", according to the POSIX
325 * definition in 2.2.2.52. Orphaned process groups are not to be affected
326 * by terminal-generated stop signals. Newly orphaned process groups are
327 * to receive a SIGHUP and a SIGCONT.
329 * "I ask you, have you ever known what it is to be an orphan?"
331 static int will_become_orphaned_pgrp(struct pid *pgrp,
332 struct task_struct *ignored_task)
334 struct task_struct *p;
336 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
337 if ((p == ignored_task) ||
338 (p->exit_state && thread_group_empty(p)) ||
339 is_global_init(p->real_parent))
342 if (task_pgrp(p->real_parent) != pgrp &&
343 task_session(p->real_parent) == task_session(p))
345 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
350 int is_current_pgrp_orphaned(void)
354 read_lock(&tasklist_lock);
355 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
356 read_unlock(&tasklist_lock);
361 static bool has_stopped_jobs(struct pid *pgrp)
363 struct task_struct *p;
365 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
366 if (p->signal->flags & SIGNAL_STOP_STOPPED)
368 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
374 * Check to see if any process groups have become orphaned as
375 * a result of our exiting, and if they have any stopped jobs,
376 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
379 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
381 struct pid *pgrp = task_pgrp(tsk);
382 struct task_struct *ignored_task = tsk;
385 /* exit: our father is in a different pgrp than
386 * we are and we were the only connection outside.
388 parent = tsk->real_parent;
390 /* reparent: our child is in a different pgrp than
391 * we are, and it was the only connection outside.
395 if (task_pgrp(parent) != pgrp &&
396 task_session(parent) == task_session(tsk) &&
397 will_become_orphaned_pgrp(pgrp, ignored_task) &&
398 has_stopped_jobs(pgrp)) {
399 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
400 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
406 * A task is exiting. If it owned this mm, find a new owner for the mm.
408 void mm_update_next_owner(struct mm_struct *mm)
410 struct task_struct *c, *g, *p = current;
414 * If the exiting or execing task is not the owner, it's
415 * someone else's problem.
420 * The current owner is exiting/execing and there are no other
421 * candidates. Do not leave the mm pointing to a possibly
422 * freed task structure.
424 if (atomic_read(&mm->mm_users) <= 1) {
429 read_lock(&tasklist_lock);
431 * Search in the children
433 list_for_each_entry(c, &p->children, sibling) {
435 goto assign_new_owner;
439 * Search in the siblings
441 list_for_each_entry(c, &p->real_parent->children, sibling) {
443 goto assign_new_owner;
447 * Search through everything else, we should not get here often.
449 for_each_process(g) {
450 if (g->flags & PF_KTHREAD)
452 for_each_thread(g, c) {
454 goto assign_new_owner;
459 read_unlock(&tasklist_lock);
461 * We found no owner yet mm_users > 1: this implies that we are
462 * most likely racing with swapoff (try_to_unuse()) or /proc or
463 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
472 * The task_lock protects c->mm from changing.
473 * We always want mm->owner->mm == mm
477 * Delay read_unlock() till we have the task_lock()
478 * to ensure that c does not slip away underneath us
480 read_unlock(&tasklist_lock);
490 #endif /* CONFIG_MEMCG */
493 * Turn us into a lazy TLB process if we
496 static void exit_mm(void)
498 struct mm_struct *mm = current->mm;
499 struct core_state *core_state;
501 exit_mm_release(current, mm);
506 * Serialize with any possible pending coredump.
507 * We must hold mmap_sem around checking core_state
508 * and clearing tsk->mm. The core-inducing thread
509 * will increment ->nr_threads for each thread in the
510 * group with ->mm != NULL.
512 down_read(&mm->mmap_sem);
513 core_state = mm->core_state;
515 struct core_thread self;
517 up_read(&mm->mmap_sem);
520 if (self.task->flags & PF_SIGNALED)
521 self.next = xchg(&core_state->dumper.next, &self);
525 * Implies mb(), the result of xchg() must be visible
526 * to core_state->dumper.
528 if (atomic_dec_and_test(&core_state->nr_threads))
529 complete(&core_state->startup);
532 set_current_state(TASK_UNINTERRUPTIBLE);
533 if (!self.task) /* see coredump_finish() */
535 freezable_schedule();
537 __set_current_state(TASK_RUNNING);
538 down_read(&mm->mmap_sem);
541 BUG_ON(mm != current->active_mm);
542 /* more a memory barrier than a real lock */
545 up_read(&mm->mmap_sem);
546 enter_lazy_tlb(mm, current);
547 task_unlock(current);
548 mm_update_next_owner(mm);
550 if (test_thread_flag(TIF_MEMDIE))
554 static struct task_struct *find_alive_thread(struct task_struct *p)
556 struct task_struct *t;
558 for_each_thread(p, t) {
559 if (!(t->flags & PF_EXITING))
565 static struct task_struct *find_child_reaper(struct task_struct *father,
566 struct list_head *dead)
567 __releases(&tasklist_lock)
568 __acquires(&tasklist_lock)
570 struct pid_namespace *pid_ns = task_active_pid_ns(father);
571 struct task_struct *reaper = pid_ns->child_reaper;
572 struct task_struct *p, *n;
574 if (likely(reaper != father))
577 reaper = find_alive_thread(father);
579 pid_ns->child_reaper = reaper;
583 write_unlock_irq(&tasklist_lock);
585 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
586 list_del_init(&p->ptrace_entry);
590 zap_pid_ns_processes(pid_ns);
591 write_lock_irq(&tasklist_lock);
597 * When we die, we re-parent all our children, and try to:
598 * 1. give them to another thread in our thread group, if such a member exists
599 * 2. give it to the first ancestor process which prctl'd itself as a
600 * child_subreaper for its children (like a service manager)
601 * 3. give it to the init process (PID 1) in our pid namespace
603 static struct task_struct *find_new_reaper(struct task_struct *father,
604 struct task_struct *child_reaper)
606 struct task_struct *thread, *reaper;
608 thread = find_alive_thread(father);
612 if (father->signal->has_child_subreaper) {
613 unsigned int ns_level = task_pid(father)->level;
615 * Find the first ->is_child_subreaper ancestor in our pid_ns.
616 * We can't check reaper != child_reaper to ensure we do not
617 * cross the namespaces, the exiting parent could be injected
618 * by setns() + fork().
619 * We check pid->level, this is slightly more efficient than
620 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
622 for (reaper = father->real_parent;
623 task_pid(reaper)->level == ns_level;
624 reaper = reaper->real_parent) {
625 if (reaper == &init_task)
627 if (!reaper->signal->is_child_subreaper)
629 thread = find_alive_thread(reaper);
639 * Any that need to be release_task'd are put on the @dead list.
641 static void reparent_leader(struct task_struct *father, struct task_struct *p,
642 struct list_head *dead)
644 if (unlikely(p->exit_state == EXIT_DEAD))
647 /* We don't want people slaying init. */
648 p->exit_signal = SIGCHLD;
650 /* If it has exited notify the new parent about this child's death. */
652 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
653 if (do_notify_parent(p, p->exit_signal)) {
654 p->exit_state = EXIT_DEAD;
655 list_add(&p->ptrace_entry, dead);
659 kill_orphaned_pgrp(p, father);
663 * This does two things:
665 * A. Make init inherit all the child processes
666 * B. Check to see if any process groups have become orphaned
667 * as a result of our exiting, and if they have any stopped
668 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
670 static void forget_original_parent(struct task_struct *father,
671 struct list_head *dead)
673 struct task_struct *p, *t, *reaper;
675 if (unlikely(!list_empty(&father->ptraced)))
676 exit_ptrace(father, dead);
678 /* Can drop and reacquire tasklist_lock */
679 reaper = find_child_reaper(father, dead);
680 if (list_empty(&father->children))
683 reaper = find_new_reaper(father, reaper);
684 list_for_each_entry(p, &father->children, sibling) {
685 for_each_thread(p, t) {
686 t->real_parent = reaper;
687 BUG_ON((!t->ptrace) != (t->parent == father));
688 if (likely(!t->ptrace))
689 t->parent = t->real_parent;
690 if (t->pdeath_signal)
691 group_send_sig_info(t->pdeath_signal,
696 * If this is a threaded reparent there is no need to
697 * notify anyone anything has happened.
699 if (!same_thread_group(reaper, father))
700 reparent_leader(father, p, dead);
702 list_splice_tail_init(&father->children, &reaper->children);
706 * Send signals to all our closest relatives so that they know
707 * to properly mourn us..
709 static void exit_notify(struct task_struct *tsk, int group_dead)
712 struct task_struct *p, *n;
715 write_lock_irq(&tasklist_lock);
716 forget_original_parent(tsk, &dead);
719 kill_orphaned_pgrp(tsk->group_leader, NULL);
721 if (unlikely(tsk->ptrace)) {
722 int sig = thread_group_leader(tsk) &&
723 thread_group_empty(tsk) &&
724 !ptrace_reparented(tsk) ?
725 tsk->exit_signal : SIGCHLD;
726 autoreap = do_notify_parent(tsk, sig);
727 } else if (thread_group_leader(tsk)) {
728 autoreap = thread_group_empty(tsk) &&
729 do_notify_parent(tsk, tsk->exit_signal);
734 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
735 if (tsk->exit_state == EXIT_DEAD)
736 list_add(&tsk->ptrace_entry, &dead);
738 /* mt-exec, de_thread() is waiting for group leader */
739 if (unlikely(tsk->signal->notify_count < 0))
740 wake_up_process(tsk->signal->group_exit_task);
741 write_unlock_irq(&tasklist_lock);
743 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
744 list_del_init(&p->ptrace_entry);
749 #ifdef CONFIG_DEBUG_STACK_USAGE
750 static void check_stack_usage(void)
752 static DEFINE_SPINLOCK(low_water_lock);
753 static int lowest_to_date = THREAD_SIZE;
756 free = stack_not_used(current);
758 if (free >= lowest_to_date)
761 spin_lock(&low_water_lock);
762 if (free < lowest_to_date) {
763 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
764 current->comm, task_pid_nr(current), free);
765 lowest_to_date = free;
767 spin_unlock(&low_water_lock);
770 static inline void check_stack_usage(void) {}
773 void __noreturn do_exit(long code)
775 struct task_struct *tsk = current;
779 * We can get here from a kernel oops, sometimes with preemption off.
780 * Start by checking for critical errors.
781 * Then fix up important state like USER_DS and preemption.
782 * Then do everything else.
785 WARN_ON(blk_needs_flush_plug(tsk));
787 if (unlikely(in_interrupt()))
788 panic("Aiee, killing interrupt handler!");
789 if (unlikely(!tsk->pid))
790 panic("Attempted to kill the idle task!");
793 * If do_exit is called because this processes oopsed, it's possible
794 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
795 * continuing. Amongst other possible reasons, this is to prevent
796 * mm_release()->clear_child_tid() from writing to a user-controlled
801 if (unlikely(in_atomic())) {
802 pr_info("note: %s[%d] exited with preempt_count %d\n",
803 current->comm, task_pid_nr(current),
805 preempt_count_set(PREEMPT_ENABLED);
808 profile_task_exit(tsk);
811 ptrace_event(PTRACE_EVENT_EXIT, code);
813 validate_creds_for_do_exit(tsk);
816 * We're taking recursive faults here in do_exit. Safest is to just
817 * leave this task alone and wait for reboot.
819 if (unlikely(tsk->flags & PF_EXITING)) {
820 pr_alert("Fixing recursive fault but reboot is needed!\n");
821 futex_exit_recursive(tsk);
822 set_current_state(TASK_UNINTERRUPTIBLE);
826 exit_signals(tsk); /* sets PF_EXITING */
828 /* sync mm's RSS info before statistics gathering */
830 sync_mm_rss(tsk->mm);
831 acct_update_integrals(tsk);
832 group_dead = atomic_dec_and_test(&tsk->signal->live);
835 * If the last thread of global init has exited, panic
836 * immediately to get a useable coredump.
838 if (unlikely(is_global_init(tsk)))
839 panic("Attempted to kill init! exitcode=0x%08x\n",
840 tsk->signal->group_exit_code ?: (int)code);
842 #ifdef CONFIG_POSIX_TIMERS
843 hrtimer_cancel(&tsk->signal->real_timer);
844 exit_itimers(tsk->signal);
847 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
849 acct_collect(code, group_dead);
854 tsk->exit_code = code;
855 taskstats_exit(tsk, group_dead);
861 trace_sched_process_exit(tsk);
868 disassociate_ctty(1);
869 exit_task_namespaces(tsk);
874 * Flush inherited counters to the parent - before the parent
875 * gets woken up by child-exit notifications.
877 * because of cgroup mode, must be called before cgroup_exit()
879 perf_event_exit_task(tsk);
881 sched_autogroup_exit_task(tsk);
885 * FIXME: do that only when needed, using sched_exit tracepoint
887 flush_ptrace_hw_breakpoint(tsk);
889 exit_tasks_rcu_start();
890 exit_notify(tsk, group_dead);
891 proc_exit_connector(tsk);
892 mpol_put_task_policy(tsk);
894 if (unlikely(current->pi_state_cache))
895 kfree(current->pi_state_cache);
898 * Make sure we are holding no locks:
900 debug_check_no_locks_held();
903 exit_io_context(tsk);
905 if (tsk->splice_pipe)
906 free_pipe_info(tsk->splice_pipe);
908 if (tsk->task_frag.page)
909 put_page(tsk->task_frag.page);
911 validate_creds_for_do_exit(tsk);
916 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
918 exit_tasks_rcu_finish();
920 lockdep_free_task(tsk);
923 EXPORT_SYMBOL_GPL(do_exit);
925 void complete_and_exit(struct completion *comp, long code)
932 EXPORT_SYMBOL(complete_and_exit);
934 SYSCALL_DEFINE1(exit, int, error_code)
936 do_exit((error_code&0xff)<<8);
940 * Take down every thread in the group. This is called by fatal signals
941 * as well as by sys_exit_group (below).
944 do_group_exit(int exit_code)
946 struct signal_struct *sig = current->signal;
948 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
950 if (signal_group_exit(sig))
951 exit_code = sig->group_exit_code;
952 else if (!thread_group_empty(current)) {
953 struct sighand_struct *const sighand = current->sighand;
955 spin_lock_irq(&sighand->siglock);
956 if (signal_group_exit(sig))
957 /* Another thread got here before we took the lock. */
958 exit_code = sig->group_exit_code;
960 sig->group_exit_code = exit_code;
961 sig->flags = SIGNAL_GROUP_EXIT;
962 zap_other_threads(current);
964 spin_unlock_irq(&sighand->siglock);
972 * this kills every thread in the thread group. Note that any externally
973 * wait4()-ing process will get the correct exit code - even if this
974 * thread is not the thread group leader.
976 SYSCALL_DEFINE1(exit_group, int, error_code)
978 do_group_exit((error_code & 0xff) << 8);
991 enum pid_type wo_type;
995 struct waitid_info *wo_info;
997 struct rusage *wo_rusage;
999 wait_queue_entry_t child_wait;
1003 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1005 return wo->wo_type == PIDTYPE_MAX ||
1006 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1010 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1012 if (!eligible_pid(wo, p))
1016 * Wait for all children (clone and not) if __WALL is set or
1017 * if it is traced by us.
1019 if (ptrace || (wo->wo_flags & __WALL))
1023 * Otherwise, wait for clone children *only* if __WCLONE is set;
1024 * otherwise, wait for non-clone children *only*.
1026 * Note: a "clone" child here is one that reports to its parent
1027 * using a signal other than SIGCHLD, or a non-leader thread which
1028 * we can only see if it is traced by us.
1030 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1037 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1038 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1039 * the lock and this task is uninteresting. If we return nonzero, we have
1040 * released the lock and the system call should return.
1042 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1045 pid_t pid = task_pid_vnr(p);
1046 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1047 struct waitid_info *infop;
1049 if (!likely(wo->wo_flags & WEXITED))
1052 if (unlikely(wo->wo_flags & WNOWAIT)) {
1053 status = p->exit_code;
1055 read_unlock(&tasklist_lock);
1056 sched_annotate_sleep();
1058 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1063 * Move the task's state to DEAD/TRACE, only one thread can do this.
1065 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1066 EXIT_TRACE : EXIT_DEAD;
1067 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1070 * We own this thread, nobody else can reap it.
1072 read_unlock(&tasklist_lock);
1073 sched_annotate_sleep();
1076 * Check thread_group_leader() to exclude the traced sub-threads.
1078 if (state == EXIT_DEAD && thread_group_leader(p)) {
1079 struct signal_struct *sig = p->signal;
1080 struct signal_struct *psig = current->signal;
1081 unsigned long maxrss;
1082 u64 tgutime, tgstime;
1085 * The resource counters for the group leader are in its
1086 * own task_struct. Those for dead threads in the group
1087 * are in its signal_struct, as are those for the child
1088 * processes it has previously reaped. All these
1089 * accumulate in the parent's signal_struct c* fields.
1091 * We don't bother to take a lock here to protect these
1092 * p->signal fields because the whole thread group is dead
1093 * and nobody can change them.
1095 * psig->stats_lock also protects us from our sub-theads
1096 * which can reap other children at the same time. Until
1097 * we change k_getrusage()-like users to rely on this lock
1098 * we have to take ->siglock as well.
1100 * We use thread_group_cputime_adjusted() to get times for
1101 * the thread group, which consolidates times for all threads
1102 * in the group including the group leader.
1104 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1105 spin_lock_irq(¤t->sighand->siglock);
1106 write_seqlock(&psig->stats_lock);
1107 psig->cutime += tgutime + sig->cutime;
1108 psig->cstime += tgstime + sig->cstime;
1109 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1111 p->min_flt + sig->min_flt + sig->cmin_flt;
1113 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1115 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1117 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1119 task_io_get_inblock(p) +
1120 sig->inblock + sig->cinblock;
1122 task_io_get_oublock(p) +
1123 sig->oublock + sig->coublock;
1124 maxrss = max(sig->maxrss, sig->cmaxrss);
1125 if (psig->cmaxrss < maxrss)
1126 psig->cmaxrss = maxrss;
1127 task_io_accounting_add(&psig->ioac, &p->ioac);
1128 task_io_accounting_add(&psig->ioac, &sig->ioac);
1129 write_sequnlock(&psig->stats_lock);
1130 spin_unlock_irq(¤t->sighand->siglock);
1134 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1135 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1136 ? p->signal->group_exit_code : p->exit_code;
1137 wo->wo_stat = status;
1139 if (state == EXIT_TRACE) {
1140 write_lock_irq(&tasklist_lock);
1141 /* We dropped tasklist, ptracer could die and untrace */
1144 /* If parent wants a zombie, don't release it now */
1145 state = EXIT_ZOMBIE;
1146 if (do_notify_parent(p, p->exit_signal))
1148 p->exit_state = state;
1149 write_unlock_irq(&tasklist_lock);
1151 if (state == EXIT_DEAD)
1155 infop = wo->wo_info;
1157 if ((status & 0x7f) == 0) {
1158 infop->cause = CLD_EXITED;
1159 infop->status = status >> 8;
1161 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1162 infop->status = status & 0x7f;
1171 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1174 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1175 return &p->exit_code;
1177 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1178 return &p->signal->group_exit_code;
1184 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1186 * @ptrace: is the wait for ptrace
1187 * @p: task to wait for
1189 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1192 * read_lock(&tasklist_lock), which is released if return value is
1193 * non-zero. Also, grabs and releases @p->sighand->siglock.
1196 * 0 if wait condition didn't exist and search for other wait conditions
1197 * should continue. Non-zero return, -errno on failure and @p's pid on
1198 * success, implies that tasklist_lock is released and wait condition
1199 * search should terminate.
1201 static int wait_task_stopped(struct wait_opts *wo,
1202 int ptrace, struct task_struct *p)
1204 struct waitid_info *infop;
1205 int exit_code, *p_code, why;
1206 uid_t uid = 0; /* unneeded, required by compiler */
1210 * Traditionally we see ptrace'd stopped tasks regardless of options.
1212 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1215 if (!task_stopped_code(p, ptrace))
1219 spin_lock_irq(&p->sighand->siglock);
1221 p_code = task_stopped_code(p, ptrace);
1222 if (unlikely(!p_code))
1225 exit_code = *p_code;
1229 if (!unlikely(wo->wo_flags & WNOWAIT))
1232 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1234 spin_unlock_irq(&p->sighand->siglock);
1239 * Now we are pretty sure this task is interesting.
1240 * Make sure it doesn't get reaped out from under us while we
1241 * give up the lock and then examine it below. We don't want to
1242 * keep holding onto the tasklist_lock while we call getrusage and
1243 * possibly take page faults for user memory.
1246 pid = task_pid_vnr(p);
1247 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1248 read_unlock(&tasklist_lock);
1249 sched_annotate_sleep();
1251 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1254 if (likely(!(wo->wo_flags & WNOWAIT)))
1255 wo->wo_stat = (exit_code << 8) | 0x7f;
1257 infop = wo->wo_info;
1260 infop->status = exit_code;
1268 * Handle do_wait work for one task in a live, non-stopped state.
1269 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1270 * the lock and this task is uninteresting. If we return nonzero, we have
1271 * released the lock and the system call should return.
1273 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1275 struct waitid_info *infop;
1279 if (!unlikely(wo->wo_flags & WCONTINUED))
1282 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1285 spin_lock_irq(&p->sighand->siglock);
1286 /* Re-check with the lock held. */
1287 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1288 spin_unlock_irq(&p->sighand->siglock);
1291 if (!unlikely(wo->wo_flags & WNOWAIT))
1292 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1293 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1294 spin_unlock_irq(&p->sighand->siglock);
1296 pid = task_pid_vnr(p);
1298 read_unlock(&tasklist_lock);
1299 sched_annotate_sleep();
1301 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1304 infop = wo->wo_info;
1306 wo->wo_stat = 0xffff;
1308 infop->cause = CLD_CONTINUED;
1311 infop->status = SIGCONT;
1317 * Consider @p for a wait by @parent.
1319 * -ECHILD should be in ->notask_error before the first call.
1320 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1321 * Returns zero if the search for a child should continue;
1322 * then ->notask_error is 0 if @p is an eligible child,
1325 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1326 struct task_struct *p)
1329 * We can race with wait_task_zombie() from another thread.
1330 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1331 * can't confuse the checks below.
1333 int exit_state = READ_ONCE(p->exit_state);
1336 if (unlikely(exit_state == EXIT_DEAD))
1339 ret = eligible_child(wo, ptrace, p);
1343 if (unlikely(exit_state == EXIT_TRACE)) {
1345 * ptrace == 0 means we are the natural parent. In this case
1346 * we should clear notask_error, debugger will notify us.
1348 if (likely(!ptrace))
1349 wo->notask_error = 0;
1353 if (likely(!ptrace) && unlikely(p->ptrace)) {
1355 * If it is traced by its real parent's group, just pretend
1356 * the caller is ptrace_do_wait() and reap this child if it
1359 * This also hides group stop state from real parent; otherwise
1360 * a single stop can be reported twice as group and ptrace stop.
1361 * If a ptracer wants to distinguish these two events for its
1362 * own children it should create a separate process which takes
1363 * the role of real parent.
1365 if (!ptrace_reparented(p))
1370 if (exit_state == EXIT_ZOMBIE) {
1371 /* we don't reap group leaders with subthreads */
1372 if (!delay_group_leader(p)) {
1374 * A zombie ptracee is only visible to its ptracer.
1375 * Notification and reaping will be cascaded to the
1376 * real parent when the ptracer detaches.
1378 if (unlikely(ptrace) || likely(!p->ptrace))
1379 return wait_task_zombie(wo, p);
1383 * Allow access to stopped/continued state via zombie by
1384 * falling through. Clearing of notask_error is complex.
1388 * If WEXITED is set, notask_error should naturally be
1389 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1390 * so, if there are live subthreads, there are events to
1391 * wait for. If all subthreads are dead, it's still safe
1392 * to clear - this function will be called again in finite
1393 * amount time once all the subthreads are released and
1394 * will then return without clearing.
1398 * Stopped state is per-task and thus can't change once the
1399 * target task dies. Only continued and exited can happen.
1400 * Clear notask_error if WCONTINUED | WEXITED.
1402 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1403 wo->notask_error = 0;
1406 * @p is alive and it's gonna stop, continue or exit, so
1407 * there always is something to wait for.
1409 wo->notask_error = 0;
1413 * Wait for stopped. Depending on @ptrace, different stopped state
1414 * is used and the two don't interact with each other.
1416 ret = wait_task_stopped(wo, ptrace, p);
1421 * Wait for continued. There's only one continued state and the
1422 * ptracer can consume it which can confuse the real parent. Don't
1423 * use WCONTINUED from ptracer. You don't need or want it.
1425 return wait_task_continued(wo, p);
1429 * Do the work of do_wait() for one thread in the group, @tsk.
1431 * -ECHILD should be in ->notask_error before the first call.
1432 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1433 * Returns zero if the search for a child should continue; then
1434 * ->notask_error is 0 if there were any eligible children,
1437 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1439 struct task_struct *p;
1441 list_for_each_entry(p, &tsk->children, sibling) {
1442 int ret = wait_consider_task(wo, 0, p);
1451 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1453 struct task_struct *p;
1455 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1456 int ret = wait_consider_task(wo, 1, p);
1465 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1466 int sync, void *key)
1468 struct wait_opts *wo = container_of(wait, struct wait_opts,
1470 struct task_struct *p = key;
1472 if (!eligible_pid(wo, p))
1475 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1478 return default_wake_function(wait, mode, sync, key);
1481 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1483 __wake_up_sync_key(&parent->signal->wait_chldexit,
1484 TASK_INTERRUPTIBLE, 1, p);
1487 static long do_wait(struct wait_opts *wo)
1489 struct task_struct *tsk;
1492 trace_sched_process_wait(wo->wo_pid);
1494 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1495 wo->child_wait.private = current;
1496 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1499 * If there is nothing that can match our criteria, just get out.
1500 * We will clear ->notask_error to zero if we see any child that
1501 * might later match our criteria, even if we are not able to reap
1504 wo->notask_error = -ECHILD;
1505 if ((wo->wo_type < PIDTYPE_MAX) &&
1506 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1509 set_current_state(TASK_INTERRUPTIBLE);
1510 read_lock(&tasklist_lock);
1513 retval = do_wait_thread(wo, tsk);
1517 retval = ptrace_do_wait(wo, tsk);
1521 if (wo->wo_flags & __WNOTHREAD)
1523 } while_each_thread(current, tsk);
1524 read_unlock(&tasklist_lock);
1527 retval = wo->notask_error;
1528 if (!retval && !(wo->wo_flags & WNOHANG)) {
1529 retval = -ERESTARTSYS;
1530 if (!signal_pending(current)) {
1536 __set_current_state(TASK_RUNNING);
1537 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1541 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1542 int options, struct rusage *ru)
1544 struct wait_opts wo;
1545 struct pid *pid = NULL;
1549 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1550 __WNOTHREAD|__WCLONE|__WALL))
1552 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1565 type = PIDTYPE_PGID;
1573 if (type < PIDTYPE_MAX)
1574 pid = find_get_pid(upid);
1578 wo.wo_flags = options;
1587 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1588 infop, int, options, struct rusage __user *, ru)
1591 struct waitid_info info = {.status = 0};
1592 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1598 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1604 if (!user_access_begin(VERIFY_WRITE, infop, sizeof(*infop)))
1607 unsafe_put_user(signo, &infop->si_signo, Efault);
1608 unsafe_put_user(0, &infop->si_errno, Efault);
1609 unsafe_put_user(info.cause, &infop->si_code, Efault);
1610 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1611 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1612 unsafe_put_user(info.status, &infop->si_status, Efault);
1620 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1623 struct wait_opts wo;
1624 struct pid *pid = NULL;
1628 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1629 __WNOTHREAD|__WCLONE|__WALL))
1632 /* -INT_MIN is not defined */
1633 if (upid == INT_MIN)
1638 else if (upid < 0) {
1639 type = PIDTYPE_PGID;
1640 pid = find_get_pid(-upid);
1641 } else if (upid == 0) {
1642 type = PIDTYPE_PGID;
1643 pid = get_task_pid(current, PIDTYPE_PGID);
1644 } else /* upid > 0 */ {
1646 pid = find_get_pid(upid);
1651 wo.wo_flags = options | WEXITED;
1657 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1663 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1664 int, options, struct rusage __user *, ru)
1667 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1670 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1676 #ifdef __ARCH_WANT_SYS_WAITPID
1679 * sys_waitpid() remains for compatibility. waitpid() should be
1680 * implemented by calling sys_wait4() from libc.a.
1682 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1684 return kernel_wait4(pid, stat_addr, options, NULL);
1689 #ifdef CONFIG_COMPAT
1690 COMPAT_SYSCALL_DEFINE4(wait4,
1692 compat_uint_t __user *, stat_addr,
1694 struct compat_rusage __user *, ru)
1697 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1699 if (ru && put_compat_rusage(&r, ru))
1705 COMPAT_SYSCALL_DEFINE5(waitid,
1706 int, which, compat_pid_t, pid,
1707 struct compat_siginfo __user *, infop, int, options,
1708 struct compat_rusage __user *, uru)
1711 struct waitid_info info = {.status = 0};
1712 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1718 /* kernel_waitid() overwrites everything in ru */
1719 if (COMPAT_USE_64BIT_TIME)
1720 err = copy_to_user(uru, &ru, sizeof(ru));
1722 err = put_compat_rusage(&ru, uru);
1731 if (!user_access_begin(VERIFY_WRITE, infop, sizeof(*infop)))
1734 unsafe_put_user(signo, &infop->si_signo, Efault);
1735 unsafe_put_user(0, &infop->si_errno, Efault);
1736 unsafe_put_user(info.cause, &infop->si_code, Efault);
1737 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1738 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1739 unsafe_put_user(info.status, &infop->si_status, Efault);
1748 __weak void abort(void)
1752 /* if that doesn't kill us, halt */
1753 panic("Oops failed to kill thread");
1755 EXPORT_SYMBOL(abort);