4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/sched/autogroup.h>
10 #include <linux/interrupt.h>
11 #include <linux/module.h>
12 #include <linux/capability.h>
13 #include <linux/completion.h>
14 #include <linux/personality.h>
15 #include <linux/tty.h>
16 #include <linux/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/fdtable.h>
23 #include <linux/freezer.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/cgroup.h>
36 #include <linux/syscalls.h>
37 #include <linux/signal.h>
38 #include <linux/posix-timers.h>
39 #include <linux/cn_proc.h>
40 #include <linux/mutex.h>
41 #include <linux/futex.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/tracehook.h>
48 #include <linux/fs_struct.h>
49 #include <linux/userfaultfd_k.h>
50 #include <linux/init_task.h>
51 #include <linux/perf_event.h>
52 #include <trace/events/sched.h>
53 #include <linux/hw_breakpoint.h>
54 #include <linux/oom.h>
55 #include <linux/writeback.h>
56 #include <linux/shm.h>
57 #include <linux/kcov.h>
58 #include <linux/random.h>
59 #include <linux/rcuwait.h>
61 #include <linux/uaccess.h>
62 #include <asm/unistd.h>
63 #include <asm/pgtable.h>
64 #include <asm/mmu_context.h>
66 static void __unhash_process(struct task_struct *p, bool group_dead)
69 detach_pid(p, PIDTYPE_PID);
71 detach_pid(p, PIDTYPE_PGID);
72 detach_pid(p, PIDTYPE_SID);
74 list_del_rcu(&p->tasks);
75 list_del_init(&p->sibling);
76 __this_cpu_dec(process_counts);
78 list_del_rcu(&p->thread_group);
79 list_del_rcu(&p->thread_node);
83 * This function expects the tasklist_lock write-locked.
85 static void __exit_signal(struct task_struct *tsk)
87 struct signal_struct *sig = tsk->signal;
88 bool group_dead = thread_group_leader(tsk);
89 struct sighand_struct *sighand;
90 struct tty_struct *uninitialized_var(tty);
93 sighand = rcu_dereference_check(tsk->sighand,
94 lockdep_tasklist_lock_is_held());
95 spin_lock(&sighand->siglock);
97 #ifdef CONFIG_POSIX_TIMERS
98 posix_cpu_timers_exit(tsk);
100 posix_cpu_timers_exit_group(tsk);
103 * This can only happen if the caller is de_thread().
104 * FIXME: this is the temporary hack, we should teach
105 * posix-cpu-timers to handle this case correctly.
107 if (unlikely(has_group_leader_pid(tsk)))
108 posix_cpu_timers_exit_group(tsk);
117 * If there is any task waiting for the group exit
120 if (sig->notify_count > 0 && !--sig->notify_count)
121 wake_up_process(sig->group_exit_task);
123 if (tsk == sig->curr_target)
124 sig->curr_target = next_thread(tsk);
127 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
128 sizeof(unsigned long long));
131 * Accumulate here the counters for all threads as they die. We could
132 * skip the group leader because it is the last user of signal_struct,
133 * but we want to avoid the race with thread_group_cputime() which can
134 * see the empty ->thread_head list.
136 task_cputime(tsk, &utime, &stime);
137 write_seqlock(&sig->stats_lock);
140 sig->gtime += task_gtime(tsk);
141 sig->min_flt += tsk->min_flt;
142 sig->maj_flt += tsk->maj_flt;
143 sig->nvcsw += tsk->nvcsw;
144 sig->nivcsw += tsk->nivcsw;
145 sig->inblock += task_io_get_inblock(tsk);
146 sig->oublock += task_io_get_oublock(tsk);
147 task_io_accounting_add(&sig->ioac, &tsk->ioac);
148 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
150 __unhash_process(tsk, group_dead);
151 write_sequnlock(&sig->stats_lock);
154 * Do this under ->siglock, we can race with another thread
155 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
157 flush_sigqueue(&tsk->pending);
159 spin_unlock(&sighand->siglock);
161 __cleanup_sighand(sighand);
162 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
164 flush_sigqueue(&sig->shared_pending);
169 static void delayed_put_task_struct(struct rcu_head *rhp)
171 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
173 perf_event_delayed_put(tsk);
174 trace_sched_process_free(tsk);
175 put_task_struct(tsk);
179 void release_task(struct task_struct *p)
181 struct task_struct *leader;
184 /* don't need to get the RCU readlock here - the process is dead and
185 * can't be modifying its own credentials. But shut RCU-lockdep up */
187 atomic_dec(&__task_cred(p)->user->processes);
192 write_lock_irq(&tasklist_lock);
193 ptrace_release_task(p);
197 * If we are the last non-leader member of the thread
198 * group, and the leader is zombie, then notify the
199 * group leader's parent process. (if it wants notification.)
202 leader = p->group_leader;
203 if (leader != p && thread_group_empty(leader)
204 && leader->exit_state == EXIT_ZOMBIE) {
206 * If we were the last child thread and the leader has
207 * exited already, and the leader's parent ignores SIGCHLD,
208 * then we are the one who should release the leader.
210 zap_leader = do_notify_parent(leader, leader->exit_signal);
212 leader->exit_state = EXIT_DEAD;
215 write_unlock_irq(&tasklist_lock);
217 call_rcu(&p->rcu, delayed_put_task_struct);
220 if (unlikely(zap_leader))
225 * Note that if this function returns a valid task_struct pointer (!NULL)
226 * task->usage must remain >0 for the duration of the RCU critical section.
228 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
230 struct sighand_struct *sighand;
231 struct task_struct *task;
234 * We need to verify that release_task() was not called and thus
235 * delayed_put_task_struct() can't run and drop the last reference
236 * before rcu_read_unlock(). We check task->sighand != NULL,
237 * but we can read the already freed and reused memory.
240 task = rcu_dereference(*ptask);
244 probe_kernel_address(&task->sighand, sighand);
247 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
248 * was already freed we can not miss the preceding update of this
252 if (unlikely(task != READ_ONCE(*ptask)))
256 * We've re-checked that "task == *ptask", now we have two different
259 * 1. This is actually the same task/task_struct. In this case
260 * sighand != NULL tells us it is still alive.
262 * 2. This is another task which got the same memory for task_struct.
263 * We can't know this of course, and we can not trust
266 * In this case we actually return a random value, but this is
269 * If we return NULL - we can pretend that we actually noticed that
270 * *ptask was updated when the previous task has exited. Or pretend
271 * that probe_slab_address(&sighand) reads NULL.
273 * If we return the new task (because sighand is not NULL for any
274 * reason) - this is fine too. This (new) task can't go away before
277 * And note: We could even eliminate the false positive if re-read
278 * task->sighand once again to avoid the falsely NULL. But this case
279 * is very unlikely so we don't care.
287 void rcuwait_wake_up(struct rcuwait *w)
289 struct task_struct *task;
294 * Order condition vs @task, such that everything prior to the load
295 * of @task is visible. This is the condition as to why the user called
296 * rcuwait_trywake() in the first place. Pairs with set_current_state()
297 * barrier (A) in rcuwait_wait_event().
300 * [S] tsk = current [S] cond = true
307 * Avoid using task_rcu_dereference() magic as long as we are careful,
308 * see comment in rcuwait_wait_event() regarding ->exit_state.
310 task = rcu_dereference(w->task);
312 wake_up_process(task);
316 struct task_struct *try_get_task_struct(struct task_struct **ptask)
318 struct task_struct *task;
321 task = task_rcu_dereference(ptask);
323 get_task_struct(task);
330 * Determine if a process group is "orphaned", according to the POSIX
331 * definition in 2.2.2.52. Orphaned process groups are not to be affected
332 * by terminal-generated stop signals. Newly orphaned process groups are
333 * to receive a SIGHUP and a SIGCONT.
335 * "I ask you, have you ever known what it is to be an orphan?"
337 static int will_become_orphaned_pgrp(struct pid *pgrp,
338 struct task_struct *ignored_task)
340 struct task_struct *p;
342 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
343 if ((p == ignored_task) ||
344 (p->exit_state && thread_group_empty(p)) ||
345 is_global_init(p->real_parent))
348 if (task_pgrp(p->real_parent) != pgrp &&
349 task_session(p->real_parent) == task_session(p))
351 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
356 int is_current_pgrp_orphaned(void)
360 read_lock(&tasklist_lock);
361 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
362 read_unlock(&tasklist_lock);
367 static bool has_stopped_jobs(struct pid *pgrp)
369 struct task_struct *p;
371 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
372 if (p->signal->flags & SIGNAL_STOP_STOPPED)
374 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
380 * Check to see if any process groups have become orphaned as
381 * a result of our exiting, and if they have any stopped jobs,
382 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
385 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
387 struct pid *pgrp = task_pgrp(tsk);
388 struct task_struct *ignored_task = tsk;
391 /* exit: our father is in a different pgrp than
392 * we are and we were the only connection outside.
394 parent = tsk->real_parent;
396 /* reparent: our child is in a different pgrp than
397 * we are, and it was the only connection outside.
401 if (task_pgrp(parent) != pgrp &&
402 task_session(parent) == task_session(tsk) &&
403 will_become_orphaned_pgrp(pgrp, ignored_task) &&
404 has_stopped_jobs(pgrp)) {
405 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
406 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
412 * A task is exiting. If it owned this mm, find a new owner for the mm.
414 void mm_update_next_owner(struct mm_struct *mm)
416 struct task_struct *c, *g, *p = current;
420 * If the exiting or execing task is not the owner, it's
421 * someone else's problem.
426 * The current owner is exiting/execing and there are no other
427 * candidates. Do not leave the mm pointing to a possibly
428 * freed task structure.
430 if (atomic_read(&mm->mm_users) <= 1) {
435 read_lock(&tasklist_lock);
437 * Search in the children
439 list_for_each_entry(c, &p->children, sibling) {
441 goto assign_new_owner;
445 * Search in the siblings
447 list_for_each_entry(c, &p->real_parent->children, sibling) {
449 goto assign_new_owner;
453 * Search through everything else, we should not get here often.
455 for_each_process(g) {
456 if (g->flags & PF_KTHREAD)
458 for_each_thread(g, c) {
460 goto assign_new_owner;
465 read_unlock(&tasklist_lock);
467 * We found no owner yet mm_users > 1: this implies that we are
468 * most likely racing with swapoff (try_to_unuse()) or /proc or
469 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
478 * The task_lock protects c->mm from changing.
479 * We always want mm->owner->mm == mm
483 * Delay read_unlock() till we have the task_lock()
484 * to ensure that c does not slip away underneath us
486 read_unlock(&tasklist_lock);
496 #endif /* CONFIG_MEMCG */
499 * Turn us into a lazy TLB process if we
502 static void exit_mm(void)
504 struct mm_struct *mm = current->mm;
505 struct core_state *core_state;
507 mm_release(current, mm);
512 * Serialize with any possible pending coredump.
513 * We must hold mmap_sem around checking core_state
514 * and clearing tsk->mm. The core-inducing thread
515 * will increment ->nr_threads for each thread in the
516 * group with ->mm != NULL.
518 down_read(&mm->mmap_sem);
519 core_state = mm->core_state;
521 struct core_thread self;
523 up_read(&mm->mmap_sem);
526 self.next = xchg(&core_state->dumper.next, &self);
528 * Implies mb(), the result of xchg() must be visible
529 * to core_state->dumper.
531 if (atomic_dec_and_test(&core_state->nr_threads))
532 complete(&core_state->startup);
535 set_current_state(TASK_UNINTERRUPTIBLE);
536 if (!self.task) /* see coredump_finish() */
538 freezable_schedule();
540 __set_current_state(TASK_RUNNING);
541 down_read(&mm->mmap_sem);
544 BUG_ON(mm != current->active_mm);
545 /* more a memory barrier than a real lock */
548 up_read(&mm->mmap_sem);
549 enter_lazy_tlb(mm, current);
550 task_unlock(current);
551 mm_update_next_owner(mm);
552 userfaultfd_exit(mm);
554 if (test_thread_flag(TIF_MEMDIE))
558 static struct task_struct *find_alive_thread(struct task_struct *p)
560 struct task_struct *t;
562 for_each_thread(p, t) {
563 if (!(t->flags & PF_EXITING))
569 static struct task_struct *find_child_reaper(struct task_struct *father)
570 __releases(&tasklist_lock)
571 __acquires(&tasklist_lock)
573 struct pid_namespace *pid_ns = task_active_pid_ns(father);
574 struct task_struct *reaper = pid_ns->child_reaper;
576 if (likely(reaper != father))
579 reaper = find_alive_thread(father);
581 pid_ns->child_reaper = reaper;
585 write_unlock_irq(&tasklist_lock);
586 if (unlikely(pid_ns == &init_pid_ns)) {
587 panic("Attempted to kill init! exitcode=0x%08x\n",
588 father->signal->group_exit_code ?: father->exit_code);
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);
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,
695 * If this is a threaded reparent there is no need to
696 * notify anyone anything has happened.
698 if (!same_thread_group(reaper, father))
699 reparent_leader(father, p, dead);
701 list_splice_tail_init(&father->children, &reaper->children);
705 * Send signals to all our closest relatives so that they know
706 * to properly mourn us..
708 static void exit_notify(struct task_struct *tsk, int group_dead)
711 struct task_struct *p, *n;
714 write_lock_irq(&tasklist_lock);
715 forget_original_parent(tsk, &dead);
718 kill_orphaned_pgrp(tsk->group_leader, NULL);
720 if (unlikely(tsk->ptrace)) {
721 int sig = thread_group_leader(tsk) &&
722 thread_group_empty(tsk) &&
723 !ptrace_reparented(tsk) ?
724 tsk->exit_signal : SIGCHLD;
725 autoreap = do_notify_parent(tsk, sig);
726 } else if (thread_group_leader(tsk)) {
727 autoreap = thread_group_empty(tsk) &&
728 do_notify_parent(tsk, tsk->exit_signal);
733 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
734 if (tsk->exit_state == EXIT_DEAD)
735 list_add(&tsk->ptrace_entry, &dead);
737 /* mt-exec, de_thread() is waiting for group leader */
738 if (unlikely(tsk->signal->notify_count < 0))
739 wake_up_process(tsk->signal->group_exit_task);
740 write_unlock_irq(&tasklist_lock);
742 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
743 list_del_init(&p->ptrace_entry);
748 #ifdef CONFIG_DEBUG_STACK_USAGE
749 static void check_stack_usage(void)
751 static DEFINE_SPINLOCK(low_water_lock);
752 static int lowest_to_date = THREAD_SIZE;
755 free = stack_not_used(current);
757 if (free >= lowest_to_date)
760 spin_lock(&low_water_lock);
761 if (free < lowest_to_date) {
762 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
763 current->comm, task_pid_nr(current), free);
764 lowest_to_date = free;
766 spin_unlock(&low_water_lock);
769 static inline void check_stack_usage(void) {}
772 void __noreturn do_exit(long code)
774 struct task_struct *tsk = current;
776 TASKS_RCU(int tasks_rcu_i);
778 profile_task_exit(tsk);
781 WARN_ON(blk_needs_flush_plug(tsk));
783 if (unlikely(in_interrupt()))
784 panic("Aiee, killing interrupt handler!");
785 if (unlikely(!tsk->pid))
786 panic("Attempted to kill the idle task!");
789 * If do_exit is called because this processes oopsed, it's possible
790 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
791 * continuing. Amongst other possible reasons, this is to prevent
792 * mm_release()->clear_child_tid() from writing to a user-controlled
797 ptrace_event(PTRACE_EVENT_EXIT, code);
799 validate_creds_for_do_exit(tsk);
802 * We're taking recursive faults here in do_exit. Safest is to just
803 * leave this task alone and wait for reboot.
805 if (unlikely(tsk->flags & PF_EXITING)) {
806 pr_alert("Fixing recursive fault but reboot is needed!\n");
808 * We can do this unlocked here. The futex code uses
809 * this flag just to verify whether the pi state
810 * cleanup has been done or not. In the worst case it
811 * loops once more. We pretend that the cleanup was
812 * done as there is no way to return. Either the
813 * OWNER_DIED bit is set by now or we push the blocked
814 * task into the wait for ever nirwana as well.
816 tsk->flags |= PF_EXITPIDONE;
817 set_current_state(TASK_UNINTERRUPTIBLE);
821 exit_signals(tsk); /* sets PF_EXITING */
823 * Ensure that all new tsk->pi_lock acquisitions must observe
824 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
828 * Ensure that we must observe the pi_state in exit_mm() ->
829 * mm_release() -> exit_pi_state_list().
831 raw_spin_unlock_wait(&tsk->pi_lock);
833 if (unlikely(in_atomic())) {
834 pr_info("note: %s[%d] exited with preempt_count %d\n",
835 current->comm, task_pid_nr(current),
837 preempt_count_set(PREEMPT_ENABLED);
840 /* sync mm's RSS info before statistics gathering */
842 sync_mm_rss(tsk->mm);
843 acct_update_integrals(tsk);
844 group_dead = atomic_dec_and_test(&tsk->signal->live);
846 #ifdef CONFIG_POSIX_TIMERS
847 hrtimer_cancel(&tsk->signal->real_timer);
848 exit_itimers(tsk->signal);
851 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
853 acct_collect(code, group_dead);
858 tsk->exit_code = code;
859 taskstats_exit(tsk, group_dead);
865 trace_sched_process_exit(tsk);
872 disassociate_ctty(1);
873 exit_task_namespaces(tsk);
878 * Flush inherited counters to the parent - before the parent
879 * gets woken up by child-exit notifications.
881 * because of cgroup mode, must be called before cgroup_exit()
883 perf_event_exit_task(tsk);
885 sched_autogroup_exit_task(tsk);
889 * FIXME: do that only when needed, using sched_exit tracepoint
891 flush_ptrace_hw_breakpoint(tsk);
893 TASKS_RCU(preempt_disable());
894 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
895 TASKS_RCU(preempt_enable());
896 exit_notify(tsk, group_dead);
897 proc_exit_connector(tsk);
898 mpol_put_task_policy(tsk);
900 if (unlikely(current->pi_state_cache))
901 kfree(current->pi_state_cache);
904 * Make sure we are holding no locks:
906 debug_check_no_locks_held();
908 * We can do this unlocked here. The futex code uses this flag
909 * just to verify whether the pi state cleanup has been done
910 * or not. In the worst case it loops once more.
912 tsk->flags |= PF_EXITPIDONE;
915 exit_io_context(tsk);
917 if (tsk->splice_pipe)
918 free_pipe_info(tsk->splice_pipe);
920 if (tsk->task_frag.page)
921 put_page(tsk->task_frag.page);
923 validate_creds_for_do_exit(tsk);
928 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
930 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
934 EXPORT_SYMBOL_GPL(do_exit);
936 void complete_and_exit(struct completion *comp, long code)
943 EXPORT_SYMBOL(complete_and_exit);
945 SYSCALL_DEFINE1(exit, int, error_code)
947 do_exit((error_code&0xff)<<8);
951 * Take down every thread in the group. This is called by fatal signals
952 * as well as by sys_exit_group (below).
955 do_group_exit(int exit_code)
957 struct signal_struct *sig = current->signal;
959 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
961 if (signal_group_exit(sig))
962 exit_code = sig->group_exit_code;
963 else if (!thread_group_empty(current)) {
964 struct sighand_struct *const sighand = current->sighand;
966 spin_lock_irq(&sighand->siglock);
967 if (signal_group_exit(sig))
968 /* Another thread got here before we took the lock. */
969 exit_code = sig->group_exit_code;
971 sig->group_exit_code = exit_code;
972 sig->flags = SIGNAL_GROUP_EXIT;
973 zap_other_threads(current);
975 spin_unlock_irq(&sighand->siglock);
983 * this kills every thread in the thread group. Note that any externally
984 * wait4()-ing process will get the correct exit code - even if this
985 * thread is not the thread group leader.
987 SYSCALL_DEFINE1(exit_group, int, error_code)
989 do_group_exit((error_code & 0xff) << 8);
995 enum pid_type wo_type;
999 struct siginfo __user *wo_info;
1000 int __user *wo_stat;
1001 struct rusage __user *wo_rusage;
1003 wait_queue_t child_wait;
1008 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1010 if (type != PIDTYPE_PID)
1011 task = task->group_leader;
1012 return task->pids[type].pid;
1015 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1017 return wo->wo_type == PIDTYPE_MAX ||
1018 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1022 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1024 if (!eligible_pid(wo, p))
1028 * Wait for all children (clone and not) if __WALL is set or
1029 * if it is traced by us.
1031 if (ptrace || (wo->wo_flags & __WALL))
1035 * Otherwise, wait for clone children *only* if __WCLONE is set;
1036 * otherwise, wait for non-clone children *only*.
1038 * Note: a "clone" child here is one that reports to its parent
1039 * using a signal other than SIGCHLD, or a non-leader thread which
1040 * we can only see if it is traced by us.
1042 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1048 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1049 pid_t pid, uid_t uid, int why, int status)
1051 struct siginfo __user *infop;
1052 int retval = wo->wo_rusage
1053 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1056 infop = wo->wo_info;
1059 retval = put_user(SIGCHLD, &infop->si_signo);
1061 retval = put_user(0, &infop->si_errno);
1063 retval = put_user((short)why, &infop->si_code);
1065 retval = put_user(pid, &infop->si_pid);
1067 retval = put_user(uid, &infop->si_uid);
1069 retval = put_user(status, &infop->si_status);
1077 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1078 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1079 * the lock and this task is uninteresting. If we return nonzero, we have
1080 * released the lock and the system call should return.
1082 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1084 int state, retval, status;
1085 pid_t pid = task_pid_vnr(p);
1086 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1087 struct siginfo __user *infop;
1089 if (!likely(wo->wo_flags & WEXITED))
1092 if (unlikely(wo->wo_flags & WNOWAIT)) {
1093 int exit_code = p->exit_code;
1097 read_unlock(&tasklist_lock);
1098 sched_annotate_sleep();
1100 if ((exit_code & 0x7f) == 0) {
1102 status = exit_code >> 8;
1104 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1105 status = exit_code & 0x7f;
1107 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1110 * Move the task's state to DEAD/TRACE, only one thread can do this.
1112 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1113 EXIT_TRACE : EXIT_DEAD;
1114 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1117 * We own this thread, nobody else can reap it.
1119 read_unlock(&tasklist_lock);
1120 sched_annotate_sleep();
1123 * Check thread_group_leader() to exclude the traced sub-threads.
1125 if (state == EXIT_DEAD && thread_group_leader(p)) {
1126 struct signal_struct *sig = p->signal;
1127 struct signal_struct *psig = current->signal;
1128 unsigned long maxrss;
1129 u64 tgutime, tgstime;
1132 * The resource counters for the group leader are in its
1133 * own task_struct. Those for dead threads in the group
1134 * are in its signal_struct, as are those for the child
1135 * processes it has previously reaped. All these
1136 * accumulate in the parent's signal_struct c* fields.
1138 * We don't bother to take a lock here to protect these
1139 * p->signal fields because the whole thread group is dead
1140 * and nobody can change them.
1142 * psig->stats_lock also protects us from our sub-theads
1143 * which can reap other children at the same time. Until
1144 * we change k_getrusage()-like users to rely on this lock
1145 * we have to take ->siglock as well.
1147 * We use thread_group_cputime_adjusted() to get times for
1148 * the thread group, which consolidates times for all threads
1149 * in the group including the group leader.
1151 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1152 spin_lock_irq(¤t->sighand->siglock);
1153 write_seqlock(&psig->stats_lock);
1154 psig->cutime += tgutime + sig->cutime;
1155 psig->cstime += tgstime + sig->cstime;
1156 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1158 p->min_flt + sig->min_flt + sig->cmin_flt;
1160 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1162 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1164 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1166 task_io_get_inblock(p) +
1167 sig->inblock + sig->cinblock;
1169 task_io_get_oublock(p) +
1170 sig->oublock + sig->coublock;
1171 maxrss = max(sig->maxrss, sig->cmaxrss);
1172 if (psig->cmaxrss < maxrss)
1173 psig->cmaxrss = maxrss;
1174 task_io_accounting_add(&psig->ioac, &p->ioac);
1175 task_io_accounting_add(&psig->ioac, &sig->ioac);
1176 write_sequnlock(&psig->stats_lock);
1177 spin_unlock_irq(¤t->sighand->siglock);
1180 retval = wo->wo_rusage
1181 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1182 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1183 ? p->signal->group_exit_code : p->exit_code;
1184 if (!retval && wo->wo_stat)
1185 retval = put_user(status, wo->wo_stat);
1187 infop = wo->wo_info;
1188 if (!retval && infop)
1189 retval = put_user(SIGCHLD, &infop->si_signo);
1190 if (!retval && infop)
1191 retval = put_user(0, &infop->si_errno);
1192 if (!retval && infop) {
1195 if ((status & 0x7f) == 0) {
1199 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1202 retval = put_user((short)why, &infop->si_code);
1204 retval = put_user(status, &infop->si_status);
1206 if (!retval && infop)
1207 retval = put_user(pid, &infop->si_pid);
1208 if (!retval && infop)
1209 retval = put_user(uid, &infop->si_uid);
1213 if (state == EXIT_TRACE) {
1214 write_lock_irq(&tasklist_lock);
1215 /* We dropped tasklist, ptracer could die and untrace */
1218 /* If parent wants a zombie, don't release it now */
1219 state = EXIT_ZOMBIE;
1220 if (do_notify_parent(p, p->exit_signal))
1222 p->exit_state = state;
1223 write_unlock_irq(&tasklist_lock);
1225 if (state == EXIT_DEAD)
1231 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1234 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1235 return &p->exit_code;
1237 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1238 return &p->signal->group_exit_code;
1244 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1246 * @ptrace: is the wait for ptrace
1247 * @p: task to wait for
1249 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1252 * read_lock(&tasklist_lock), which is released if return value is
1253 * non-zero. Also, grabs and releases @p->sighand->siglock.
1256 * 0 if wait condition didn't exist and search for other wait conditions
1257 * should continue. Non-zero return, -errno on failure and @p's pid on
1258 * success, implies that tasklist_lock is released and wait condition
1259 * search should terminate.
1261 static int wait_task_stopped(struct wait_opts *wo,
1262 int ptrace, struct task_struct *p)
1264 struct siginfo __user *infop;
1265 int retval, exit_code, *p_code, why;
1266 uid_t uid = 0; /* unneeded, required by compiler */
1270 * Traditionally we see ptrace'd stopped tasks regardless of options.
1272 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1275 if (!task_stopped_code(p, ptrace))
1279 spin_lock_irq(&p->sighand->siglock);
1281 p_code = task_stopped_code(p, ptrace);
1282 if (unlikely(!p_code))
1285 exit_code = *p_code;
1289 if (!unlikely(wo->wo_flags & WNOWAIT))
1292 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1294 spin_unlock_irq(&p->sighand->siglock);
1299 * Now we are pretty sure this task is interesting.
1300 * Make sure it doesn't get reaped out from under us while we
1301 * give up the lock and then examine it below. We don't want to
1302 * keep holding onto the tasklist_lock while we call getrusage and
1303 * possibly take page faults for user memory.
1306 pid = task_pid_vnr(p);
1307 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1308 read_unlock(&tasklist_lock);
1309 sched_annotate_sleep();
1311 if (unlikely(wo->wo_flags & WNOWAIT))
1312 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1314 retval = wo->wo_rusage
1315 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1316 if (!retval && wo->wo_stat)
1317 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1319 infop = wo->wo_info;
1320 if (!retval && infop)
1321 retval = put_user(SIGCHLD, &infop->si_signo);
1322 if (!retval && infop)
1323 retval = put_user(0, &infop->si_errno);
1324 if (!retval && infop)
1325 retval = put_user((short)why, &infop->si_code);
1326 if (!retval && infop)
1327 retval = put_user(exit_code, &infop->si_status);
1328 if (!retval && infop)
1329 retval = put_user(pid, &infop->si_pid);
1330 if (!retval && infop)
1331 retval = put_user(uid, &infop->si_uid);
1341 * Handle do_wait work for one task in a live, non-stopped state.
1342 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1343 * the lock and this task is uninteresting. If we return nonzero, we have
1344 * released the lock and the system call should return.
1346 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1352 if (!unlikely(wo->wo_flags & WCONTINUED))
1355 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1358 spin_lock_irq(&p->sighand->siglock);
1359 /* Re-check with the lock held. */
1360 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1361 spin_unlock_irq(&p->sighand->siglock);
1364 if (!unlikely(wo->wo_flags & WNOWAIT))
1365 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1366 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1367 spin_unlock_irq(&p->sighand->siglock);
1369 pid = task_pid_vnr(p);
1371 read_unlock(&tasklist_lock);
1372 sched_annotate_sleep();
1375 retval = wo->wo_rusage
1376 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1378 if (!retval && wo->wo_stat)
1379 retval = put_user(0xffff, wo->wo_stat);
1383 retval = wait_noreap_copyout(wo, p, pid, uid,
1384 CLD_CONTINUED, SIGCONT);
1385 BUG_ON(retval == 0);
1392 * Consider @p for a wait by @parent.
1394 * -ECHILD should be in ->notask_error before the first call.
1395 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1396 * Returns zero if the search for a child should continue;
1397 * then ->notask_error is 0 if @p is an eligible child,
1400 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1401 struct task_struct *p)
1404 * We can race with wait_task_zombie() from another thread.
1405 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1406 * can't confuse the checks below.
1408 int exit_state = ACCESS_ONCE(p->exit_state);
1411 if (unlikely(exit_state == EXIT_DEAD))
1414 ret = eligible_child(wo, ptrace, p);
1418 if (unlikely(exit_state == EXIT_TRACE)) {
1420 * ptrace == 0 means we are the natural parent. In this case
1421 * we should clear notask_error, debugger will notify us.
1423 if (likely(!ptrace))
1424 wo->notask_error = 0;
1428 if (likely(!ptrace) && unlikely(p->ptrace)) {
1430 * If it is traced by its real parent's group, just pretend
1431 * the caller is ptrace_do_wait() and reap this child if it
1434 * This also hides group stop state from real parent; otherwise
1435 * a single stop can be reported twice as group and ptrace stop.
1436 * If a ptracer wants to distinguish these two events for its
1437 * own children it should create a separate process which takes
1438 * the role of real parent.
1440 if (!ptrace_reparented(p))
1445 if (exit_state == EXIT_ZOMBIE) {
1446 /* we don't reap group leaders with subthreads */
1447 if (!delay_group_leader(p)) {
1449 * A zombie ptracee is only visible to its ptracer.
1450 * Notification and reaping will be cascaded to the
1451 * real parent when the ptracer detaches.
1453 if (unlikely(ptrace) || likely(!p->ptrace))
1454 return wait_task_zombie(wo, p);
1458 * Allow access to stopped/continued state via zombie by
1459 * falling through. Clearing of notask_error is complex.
1463 * If WEXITED is set, notask_error should naturally be
1464 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1465 * so, if there are live subthreads, there are events to
1466 * wait for. If all subthreads are dead, it's still safe
1467 * to clear - this function will be called again in finite
1468 * amount time once all the subthreads are released and
1469 * will then return without clearing.
1473 * Stopped state is per-task and thus can't change once the
1474 * target task dies. Only continued and exited can happen.
1475 * Clear notask_error if WCONTINUED | WEXITED.
1477 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1478 wo->notask_error = 0;
1481 * @p is alive and it's gonna stop, continue or exit, so
1482 * there always is something to wait for.
1484 wo->notask_error = 0;
1488 * Wait for stopped. Depending on @ptrace, different stopped state
1489 * is used and the two don't interact with each other.
1491 ret = wait_task_stopped(wo, ptrace, p);
1496 * Wait for continued. There's only one continued state and the
1497 * ptracer can consume it which can confuse the real parent. Don't
1498 * use WCONTINUED from ptracer. You don't need or want it.
1500 return wait_task_continued(wo, p);
1504 * Do the work of do_wait() for one thread in the group, @tsk.
1506 * -ECHILD should be in ->notask_error before the first call.
1507 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1508 * Returns zero if the search for a child should continue; then
1509 * ->notask_error is 0 if there were any eligible children,
1512 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1514 struct task_struct *p;
1516 list_for_each_entry(p, &tsk->children, sibling) {
1517 int ret = wait_consider_task(wo, 0, p);
1526 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1528 struct task_struct *p;
1530 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1531 int ret = wait_consider_task(wo, 1, p);
1540 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1541 int sync, void *key)
1543 struct wait_opts *wo = container_of(wait, struct wait_opts,
1545 struct task_struct *p = key;
1547 if (!eligible_pid(wo, p))
1550 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1553 return default_wake_function(wait, mode, sync, key);
1556 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1558 __wake_up_sync_key(&parent->signal->wait_chldexit,
1559 TASK_INTERRUPTIBLE, 1, p);
1562 static long do_wait(struct wait_opts *wo)
1564 struct task_struct *tsk;
1567 trace_sched_process_wait(wo->wo_pid);
1569 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1570 wo->child_wait.private = current;
1571 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1574 * If there is nothing that can match our criteria, just get out.
1575 * We will clear ->notask_error to zero if we see any child that
1576 * might later match our criteria, even if we are not able to reap
1579 wo->notask_error = -ECHILD;
1580 if ((wo->wo_type < PIDTYPE_MAX) &&
1581 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1584 set_current_state(TASK_INTERRUPTIBLE);
1585 read_lock(&tasklist_lock);
1588 retval = do_wait_thread(wo, tsk);
1592 retval = ptrace_do_wait(wo, tsk);
1596 if (wo->wo_flags & __WNOTHREAD)
1598 } while_each_thread(current, tsk);
1599 read_unlock(&tasklist_lock);
1602 retval = wo->notask_error;
1603 if (!retval && !(wo->wo_flags & WNOHANG)) {
1604 retval = -ERESTARTSYS;
1605 if (!signal_pending(current)) {
1611 __set_current_state(TASK_RUNNING);
1612 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1616 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1617 infop, int, options, struct rusage __user *, ru)
1619 struct wait_opts wo;
1620 struct pid *pid = NULL;
1624 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1625 __WNOTHREAD|__WCLONE|__WALL))
1627 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1640 type = PIDTYPE_PGID;
1648 if (type < PIDTYPE_MAX)
1649 pid = find_get_pid(upid);
1653 wo.wo_flags = options;
1663 * For a WNOHANG return, clear out all the fields
1664 * we would set so the user can easily tell the
1668 ret = put_user(0, &infop->si_signo);
1670 ret = put_user(0, &infop->si_errno);
1672 ret = put_user(0, &infop->si_code);
1674 ret = put_user(0, &infop->si_pid);
1676 ret = put_user(0, &infop->si_uid);
1678 ret = put_user(0, &infop->si_status);
1685 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1686 int, options, struct rusage __user *, ru)
1688 struct wait_opts wo;
1689 struct pid *pid = NULL;
1693 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1694 __WNOTHREAD|__WCLONE|__WALL))
1699 else if (upid < 0) {
1700 type = PIDTYPE_PGID;
1701 pid = find_get_pid(-upid);
1702 } else if (upid == 0) {
1703 type = PIDTYPE_PGID;
1704 pid = get_task_pid(current, PIDTYPE_PGID);
1705 } else /* upid > 0 */ {
1707 pid = find_get_pid(upid);
1712 wo.wo_flags = options | WEXITED;
1714 wo.wo_stat = stat_addr;
1722 #ifdef __ARCH_WANT_SYS_WAITPID
1725 * sys_waitpid() remains for compatibility. waitpid() should be
1726 * implemented by calling sys_wait4() from libc.a.
1728 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1730 return sys_wait4(pid, stat_addr, options, NULL);