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/interrupt.h>
12 #include <linux/module.h>
13 #include <linux/capability.h>
14 #include <linux/completion.h>
15 #include <linux/personality.h>
16 #include <linux/tty.h>
17 #include <linux/iocontext.h>
18 #include <linux/key.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/freezer.h>
25 #include <linux/binfmts.h>
26 #include <linux/nsproxy.h>
27 #include <linux/pid_namespace.h>
28 #include <linux/ptrace.h>
29 #include <linux/profile.h>
30 #include <linux/mount.h>
31 #include <linux/proc_fs.h>
32 #include <linux/kthread.h>
33 #include <linux/mempolicy.h>
34 #include <linux/taskstats_kern.h>
35 #include <linux/delayacct.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <linux/fs_struct.h>
50 #include <linux/userfaultfd_k.h>
51 #include <linux/init_task.h>
52 #include <linux/perf_event.h>
53 #include <trace/events/sched.h>
54 #include <linux/hw_breakpoint.h>
55 #include <linux/oom.h>
56 #include <linux/writeback.h>
57 #include <linux/shm.h>
58 #include <linux/kcov.h>
59 #include <linux/random.h>
60 #include <linux/rcuwait.h>
62 #include <linux/uaccess.h>
63 #include <asm/unistd.h>
64 #include <asm/pgtable.h>
65 #include <asm/mmu_context.h>
67 static void __unhash_process(struct task_struct *p, bool group_dead)
70 detach_pid(p, PIDTYPE_PID);
72 detach_pid(p, PIDTYPE_PGID);
73 detach_pid(p, PIDTYPE_SID);
75 list_del_rcu(&p->tasks);
76 list_del_init(&p->sibling);
77 __this_cpu_dec(process_counts);
79 list_del_rcu(&p->thread_group);
80 list_del_rcu(&p->thread_node);
84 * This function expects the tasklist_lock write-locked.
86 static void __exit_signal(struct task_struct *tsk)
88 struct signal_struct *sig = tsk->signal;
89 bool group_dead = thread_group_leader(tsk);
90 struct sighand_struct *sighand;
91 struct tty_struct *uninitialized_var(tty);
94 sighand = rcu_dereference_check(tsk->sighand,
95 lockdep_tasklist_lock_is_held());
96 spin_lock(&sighand->siglock);
98 #ifdef CONFIG_POSIX_TIMERS
99 posix_cpu_timers_exit(tsk);
101 posix_cpu_timers_exit_group(tsk);
104 * This can only happen if the caller is de_thread().
105 * FIXME: this is the temporary hack, we should teach
106 * posix-cpu-timers to handle this case correctly.
108 if (unlikely(has_group_leader_pid(tsk)))
109 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_exit_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 perf_event_delayed_put(tsk);
175 trace_sched_process_free(tsk);
176 put_task_struct(tsk);
180 void release_task(struct task_struct *p)
182 struct task_struct *leader;
185 /* don't need to get the RCU readlock here - the process is dead and
186 * can't be modifying its own credentials. But shut RCU-lockdep up */
188 atomic_dec(&__task_cred(p)->user->processes);
193 write_lock_irq(&tasklist_lock);
194 ptrace_release_task(p);
198 * If we are the last non-leader member of the thread
199 * group, and the leader is zombie, then notify the
200 * group leader's parent process. (if it wants notification.)
203 leader = p->group_leader;
204 if (leader != p && thread_group_empty(leader)
205 && leader->exit_state == EXIT_ZOMBIE) {
207 * If we were the last child thread and the leader has
208 * exited already, and the leader's parent ignores SIGCHLD,
209 * then we are the one who should release the leader.
211 zap_leader = do_notify_parent(leader, leader->exit_signal);
213 leader->exit_state = EXIT_DEAD;
216 write_unlock_irq(&tasklist_lock);
218 call_rcu(&p->rcu, delayed_put_task_struct);
221 if (unlikely(zap_leader))
226 * Note that if this function returns a valid task_struct pointer (!NULL)
227 * task->usage must remain >0 for the duration of the RCU critical section.
229 struct task_struct *task_rcu_dereference(struct task_struct **ptask)
231 struct sighand_struct *sighand;
232 struct task_struct *task;
235 * We need to verify that release_task() was not called and thus
236 * delayed_put_task_struct() can't run and drop the last reference
237 * before rcu_read_unlock(). We check task->sighand != NULL,
238 * but we can read the already freed and reused memory.
241 task = rcu_dereference(*ptask);
245 probe_kernel_address(&task->sighand, sighand);
248 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
249 * was already freed we can not miss the preceding update of this
253 if (unlikely(task != READ_ONCE(*ptask)))
257 * We've re-checked that "task == *ptask", now we have two different
260 * 1. This is actually the same task/task_struct. In this case
261 * sighand != NULL tells us it is still alive.
263 * 2. This is another task which got the same memory for task_struct.
264 * We can't know this of course, and we can not trust
267 * In this case we actually return a random value, but this is
270 * If we return NULL - we can pretend that we actually noticed that
271 * *ptask was updated when the previous task has exited. Or pretend
272 * that probe_slab_address(&sighand) reads NULL.
274 * If we return the new task (because sighand is not NULL for any
275 * reason) - this is fine too. This (new) task can't go away before
278 * And note: We could even eliminate the false positive if re-read
279 * task->sighand once again to avoid the falsely NULL. But this case
280 * is very unlikely so we don't care.
288 void rcuwait_wake_up(struct rcuwait *w)
290 struct task_struct *task;
295 * Order condition vs @task, such that everything prior to the load
296 * of @task is visible. This is the condition as to why the user called
297 * rcuwait_trywake() in the first place. Pairs with set_current_state()
298 * barrier (A) in rcuwait_wait_event().
301 * [S] tsk = current [S] cond = true
308 * Avoid using task_rcu_dereference() magic as long as we are careful,
309 * see comment in rcuwait_wait_event() regarding ->exit_state.
311 task = rcu_dereference(w->task);
313 wake_up_process(task);
317 struct task_struct *try_get_task_struct(struct task_struct **ptask)
319 struct task_struct *task;
322 task = task_rcu_dereference(ptask);
324 get_task_struct(task);
331 * Determine if a process group is "orphaned", according to the POSIX
332 * definition in 2.2.2.52. Orphaned process groups are not to be affected
333 * by terminal-generated stop signals. Newly orphaned process groups are
334 * to receive a SIGHUP and a SIGCONT.
336 * "I ask you, have you ever known what it is to be an orphan?"
338 static int will_become_orphaned_pgrp(struct pid *pgrp,
339 struct task_struct *ignored_task)
341 struct task_struct *p;
343 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
344 if ((p == ignored_task) ||
345 (p->exit_state && thread_group_empty(p)) ||
346 is_global_init(p->real_parent))
349 if (task_pgrp(p->real_parent) != pgrp &&
350 task_session(p->real_parent) == task_session(p))
352 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
357 int is_current_pgrp_orphaned(void)
361 read_lock(&tasklist_lock);
362 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
363 read_unlock(&tasklist_lock);
368 static bool has_stopped_jobs(struct pid *pgrp)
370 struct task_struct *p;
372 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
373 if (p->signal->flags & SIGNAL_STOP_STOPPED)
375 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
381 * Check to see if any process groups have become orphaned as
382 * a result of our exiting, and if they have any stopped jobs,
383 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
386 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
388 struct pid *pgrp = task_pgrp(tsk);
389 struct task_struct *ignored_task = tsk;
392 /* exit: our father is in a different pgrp than
393 * we are and we were the only connection outside.
395 parent = tsk->real_parent;
397 /* reparent: our child is in a different pgrp than
398 * we are, and it was the only connection outside.
402 if (task_pgrp(parent) != pgrp &&
403 task_session(parent) == task_session(tsk) &&
404 will_become_orphaned_pgrp(pgrp, ignored_task) &&
405 has_stopped_jobs(pgrp)) {
406 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
407 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
413 * A task is exiting. If it owned this mm, find a new owner for the mm.
415 void mm_update_next_owner(struct mm_struct *mm)
417 struct task_struct *c, *g, *p = current;
421 * If the exiting or execing task is not the owner, it's
422 * someone else's problem.
427 * The current owner is exiting/execing and there are no other
428 * candidates. Do not leave the mm pointing to a possibly
429 * freed task structure.
431 if (atomic_read(&mm->mm_users) <= 1) {
436 read_lock(&tasklist_lock);
438 * Search in the children
440 list_for_each_entry(c, &p->children, sibling) {
442 goto assign_new_owner;
446 * Search in the siblings
448 list_for_each_entry(c, &p->real_parent->children, sibling) {
450 goto assign_new_owner;
454 * Search through everything else, we should not get here often.
456 for_each_process(g) {
457 if (g->flags & PF_KTHREAD)
459 for_each_thread(g, c) {
461 goto assign_new_owner;
466 read_unlock(&tasklist_lock);
468 * We found no owner yet mm_users > 1: this implies that we are
469 * most likely racing with swapoff (try_to_unuse()) or /proc or
470 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
479 * The task_lock protects c->mm from changing.
480 * We always want mm->owner->mm == mm
484 * Delay read_unlock() till we have the task_lock()
485 * to ensure that c does not slip away underneath us
487 read_unlock(&tasklist_lock);
497 #endif /* CONFIG_MEMCG */
500 * Turn us into a lazy TLB process if we
503 static void exit_mm(void)
505 struct mm_struct *mm = current->mm;
506 struct core_state *core_state;
508 mm_release(current, mm);
513 * Serialize with any possible pending coredump.
514 * We must hold mmap_sem around checking core_state
515 * and clearing tsk->mm. The core-inducing thread
516 * will increment ->nr_threads for each thread in the
517 * group with ->mm != NULL.
519 down_read(&mm->mmap_sem);
520 core_state = mm->core_state;
522 struct core_thread self;
524 up_read(&mm->mmap_sem);
527 self.next = xchg(&core_state->dumper.next, &self);
529 * Implies mb(), the result of xchg() must be visible
530 * to core_state->dumper.
532 if (atomic_dec_and_test(&core_state->nr_threads))
533 complete(&core_state->startup);
536 set_current_state(TASK_UNINTERRUPTIBLE);
537 if (!self.task) /* see coredump_finish() */
539 freezable_schedule();
541 __set_current_state(TASK_RUNNING);
542 down_read(&mm->mmap_sem);
545 BUG_ON(mm != current->active_mm);
546 /* more a memory barrier than a real lock */
549 up_read(&mm->mmap_sem);
550 enter_lazy_tlb(mm, current);
551 task_unlock(current);
552 mm_update_next_owner(mm);
553 userfaultfd_exit(mm);
555 if (test_thread_flag(TIF_MEMDIE))
559 static struct task_struct *find_alive_thread(struct task_struct *p)
561 struct task_struct *t;
563 for_each_thread(p, t) {
564 if (!(t->flags & PF_EXITING))
570 static struct task_struct *find_child_reaper(struct task_struct *father)
571 __releases(&tasklist_lock)
572 __acquires(&tasklist_lock)
574 struct pid_namespace *pid_ns = task_active_pid_ns(father);
575 struct task_struct *reaper = pid_ns->child_reaper;
577 if (likely(reaper != father))
580 reaper = find_alive_thread(father);
582 pid_ns->child_reaper = reaper;
586 write_unlock_irq(&tasklist_lock);
587 if (unlikely(pid_ns == &init_pid_ns)) {
588 panic("Attempted to kill init! exitcode=0x%08x\n",
589 father->signal->group_exit_code ?: father->exit_code);
591 zap_pid_ns_processes(pid_ns);
592 write_lock_irq(&tasklist_lock);
598 * When we die, we re-parent all our children, and try to:
599 * 1. give them to another thread in our thread group, if such a member exists
600 * 2. give it to the first ancestor process which prctl'd itself as a
601 * child_subreaper for its children (like a service manager)
602 * 3. give it to the init process (PID 1) in our pid namespace
604 static struct task_struct *find_new_reaper(struct task_struct *father,
605 struct task_struct *child_reaper)
607 struct task_struct *thread, *reaper;
609 thread = find_alive_thread(father);
613 if (father->signal->has_child_subreaper) {
614 unsigned int ns_level = task_pid(father)->level;
616 * Find the first ->is_child_subreaper ancestor in our pid_ns.
617 * We can't check reaper != child_reaper to ensure we do not
618 * cross the namespaces, the exiting parent could be injected
619 * by setns() + fork().
620 * We check pid->level, this is slightly more efficient than
621 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
623 for (reaper = father->real_parent;
624 task_pid(reaper)->level == ns_level;
625 reaper = reaper->real_parent) {
626 if (reaper == &init_task)
628 if (!reaper->signal->is_child_subreaper)
630 thread = find_alive_thread(reaper);
640 * Any that need to be release_task'd are put on the @dead list.
642 static void reparent_leader(struct task_struct *father, struct task_struct *p,
643 struct list_head *dead)
645 if (unlikely(p->exit_state == EXIT_DEAD))
648 /* We don't want people slaying init. */
649 p->exit_signal = SIGCHLD;
651 /* If it has exited notify the new parent about this child's death. */
653 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
654 if (do_notify_parent(p, p->exit_signal)) {
655 p->exit_state = EXIT_DEAD;
656 list_add(&p->ptrace_entry, dead);
660 kill_orphaned_pgrp(p, father);
664 * This does two things:
666 * A. Make init inherit all the child processes
667 * B. Check to see if any process groups have become orphaned
668 * as a result of our exiting, and if they have any stopped
669 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
671 static void forget_original_parent(struct task_struct *father,
672 struct list_head *dead)
674 struct task_struct *p, *t, *reaper;
676 if (unlikely(!list_empty(&father->ptraced)))
677 exit_ptrace(father, dead);
679 /* Can drop and reacquire tasklist_lock */
680 reaper = find_child_reaper(father);
681 if (list_empty(&father->children))
684 reaper = find_new_reaper(father, reaper);
685 list_for_each_entry(p, &father->children, sibling) {
686 for_each_thread(p, t) {
687 t->real_parent = reaper;
688 BUG_ON((!t->ptrace) != (t->parent == father));
689 if (likely(!t->ptrace))
690 t->parent = t->real_parent;
691 if (t->pdeath_signal)
692 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;
777 TASKS_RCU(int tasks_rcu_i);
779 profile_task_exit(tsk);
782 WARN_ON(blk_needs_flush_plug(tsk));
784 if (unlikely(in_interrupt()))
785 panic("Aiee, killing interrupt handler!");
786 if (unlikely(!tsk->pid))
787 panic("Attempted to kill the idle task!");
790 * If do_exit is called because this processes oopsed, it's possible
791 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
792 * continuing. Amongst other possible reasons, this is to prevent
793 * mm_release()->clear_child_tid() from writing to a user-controlled
798 ptrace_event(PTRACE_EVENT_EXIT, code);
800 validate_creds_for_do_exit(tsk);
803 * We're taking recursive faults here in do_exit. Safest is to just
804 * leave this task alone and wait for reboot.
806 if (unlikely(tsk->flags & PF_EXITING)) {
807 pr_alert("Fixing recursive fault but reboot is needed!\n");
809 * We can do this unlocked here. The futex code uses
810 * this flag just to verify whether the pi state
811 * cleanup has been done or not. In the worst case it
812 * loops once more. We pretend that the cleanup was
813 * done as there is no way to return. Either the
814 * OWNER_DIED bit is set by now or we push the blocked
815 * task into the wait for ever nirwana as well.
817 tsk->flags |= PF_EXITPIDONE;
818 set_current_state(TASK_UNINTERRUPTIBLE);
822 exit_signals(tsk); /* sets PF_EXITING */
824 * Ensure that all new tsk->pi_lock acquisitions must observe
825 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
829 * Ensure that we must observe the pi_state in exit_mm() ->
830 * mm_release() -> exit_pi_state_list().
832 raw_spin_unlock_wait(&tsk->pi_lock);
834 if (unlikely(in_atomic())) {
835 pr_info("note: %s[%d] exited with preempt_count %d\n",
836 current->comm, task_pid_nr(current),
838 preempt_count_set(PREEMPT_ENABLED);
841 /* sync mm's RSS info before statistics gathering */
843 sync_mm_rss(tsk->mm);
844 acct_update_integrals(tsk);
845 group_dead = atomic_dec_and_test(&tsk->signal->live);
847 #ifdef CONFIG_POSIX_TIMERS
848 hrtimer_cancel(&tsk->signal->real_timer);
849 exit_itimers(tsk->signal);
852 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
854 acct_collect(code, group_dead);
859 tsk->exit_code = code;
860 taskstats_exit(tsk, group_dead);
866 trace_sched_process_exit(tsk);
873 disassociate_ctty(1);
874 exit_task_namespaces(tsk);
879 * Flush inherited counters to the parent - before the parent
880 * gets woken up by child-exit notifications.
882 * because of cgroup mode, must be called before cgroup_exit()
884 perf_event_exit_task(tsk);
886 sched_autogroup_exit_task(tsk);
890 * FIXME: do that only when needed, using sched_exit tracepoint
892 flush_ptrace_hw_breakpoint(tsk);
894 TASKS_RCU(preempt_disable());
895 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
896 TASKS_RCU(preempt_enable());
897 exit_notify(tsk, group_dead);
898 proc_exit_connector(tsk);
899 mpol_put_task_policy(tsk);
901 if (unlikely(current->pi_state_cache))
902 kfree(current->pi_state_cache);
905 * Make sure we are holding no locks:
907 debug_check_no_locks_held();
909 * We can do this unlocked here. The futex code uses this flag
910 * just to verify whether the pi state cleanup has been done
911 * or not. In the worst case it loops once more.
913 tsk->flags |= PF_EXITPIDONE;
916 exit_io_context(tsk);
918 if (tsk->splice_pipe)
919 free_pipe_info(tsk->splice_pipe);
921 if (tsk->task_frag.page)
922 put_page(tsk->task_frag.page);
924 validate_creds_for_do_exit(tsk);
929 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
931 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
935 EXPORT_SYMBOL_GPL(do_exit);
937 void complete_and_exit(struct completion *comp, long code)
944 EXPORT_SYMBOL(complete_and_exit);
946 SYSCALL_DEFINE1(exit, int, error_code)
948 do_exit((error_code&0xff)<<8);
952 * Take down every thread in the group. This is called by fatal signals
953 * as well as by sys_exit_group (below).
956 do_group_exit(int exit_code)
958 struct signal_struct *sig = current->signal;
960 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
962 if (signal_group_exit(sig))
963 exit_code = sig->group_exit_code;
964 else if (!thread_group_empty(current)) {
965 struct sighand_struct *const sighand = current->sighand;
967 spin_lock_irq(&sighand->siglock);
968 if (signal_group_exit(sig))
969 /* Another thread got here before we took the lock. */
970 exit_code = sig->group_exit_code;
972 sig->group_exit_code = exit_code;
973 sig->flags = SIGNAL_GROUP_EXIT;
974 zap_other_threads(current);
976 spin_unlock_irq(&sighand->siglock);
984 * this kills every thread in the thread group. Note that any externally
985 * wait4()-ing process will get the correct exit code - even if this
986 * thread is not the thread group leader.
988 SYSCALL_DEFINE1(exit_group, int, error_code)
990 do_group_exit((error_code & 0xff) << 8);
996 enum pid_type wo_type;
1000 struct siginfo __user *wo_info;
1001 int __user *wo_stat;
1002 struct rusage __user *wo_rusage;
1004 wait_queue_t child_wait;
1009 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1011 if (type != PIDTYPE_PID)
1012 task = task->group_leader;
1013 return task->pids[type].pid;
1016 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1018 return wo->wo_type == PIDTYPE_MAX ||
1019 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1023 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
1025 if (!eligible_pid(wo, p))
1029 * Wait for all children (clone and not) if __WALL is set or
1030 * if it is traced by us.
1032 if (ptrace || (wo->wo_flags & __WALL))
1036 * Otherwise, wait for clone children *only* if __WCLONE is set;
1037 * otherwise, wait for non-clone children *only*.
1039 * Note: a "clone" child here is one that reports to its parent
1040 * using a signal other than SIGCHLD, or a non-leader thread which
1041 * we can only see if it is traced by us.
1043 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1049 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1050 pid_t pid, uid_t uid, int why, int status)
1052 struct siginfo __user *infop;
1053 int retval = wo->wo_rusage
1054 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1057 infop = wo->wo_info;
1060 retval = put_user(SIGCHLD, &infop->si_signo);
1062 retval = put_user(0, &infop->si_errno);
1064 retval = put_user((short)why, &infop->si_code);
1066 retval = put_user(pid, &infop->si_pid);
1068 retval = put_user(uid, &infop->si_uid);
1070 retval = put_user(status, &infop->si_status);
1078 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1079 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1080 * the lock and this task is uninteresting. If we return nonzero, we have
1081 * released the lock and the system call should return.
1083 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1085 int state, retval, status;
1086 pid_t pid = task_pid_vnr(p);
1087 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1088 struct siginfo __user *infop;
1090 if (!likely(wo->wo_flags & WEXITED))
1093 if (unlikely(wo->wo_flags & WNOWAIT)) {
1094 int exit_code = p->exit_code;
1098 read_unlock(&tasklist_lock);
1099 sched_annotate_sleep();
1101 if ((exit_code & 0x7f) == 0) {
1103 status = exit_code >> 8;
1105 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1106 status = exit_code & 0x7f;
1108 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1111 * Move the task's state to DEAD/TRACE, only one thread can do this.
1113 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1114 EXIT_TRACE : EXIT_DEAD;
1115 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1118 * We own this thread, nobody else can reap it.
1120 read_unlock(&tasklist_lock);
1121 sched_annotate_sleep();
1124 * Check thread_group_leader() to exclude the traced sub-threads.
1126 if (state == EXIT_DEAD && thread_group_leader(p)) {
1127 struct signal_struct *sig = p->signal;
1128 struct signal_struct *psig = current->signal;
1129 unsigned long maxrss;
1130 u64 tgutime, tgstime;
1133 * The resource counters for the group leader are in its
1134 * own task_struct. Those for dead threads in the group
1135 * are in its signal_struct, as are those for the child
1136 * processes it has previously reaped. All these
1137 * accumulate in the parent's signal_struct c* fields.
1139 * We don't bother to take a lock here to protect these
1140 * p->signal fields because the whole thread group is dead
1141 * and nobody can change them.
1143 * psig->stats_lock also protects us from our sub-theads
1144 * which can reap other children at the same time. Until
1145 * we change k_getrusage()-like users to rely on this lock
1146 * we have to take ->siglock as well.
1148 * We use thread_group_cputime_adjusted() to get times for
1149 * the thread group, which consolidates times for all threads
1150 * in the group including the group leader.
1152 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1153 spin_lock_irq(¤t->sighand->siglock);
1154 write_seqlock(&psig->stats_lock);
1155 psig->cutime += tgutime + sig->cutime;
1156 psig->cstime += tgstime + sig->cstime;
1157 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1159 p->min_flt + sig->min_flt + sig->cmin_flt;
1161 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1163 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1165 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1167 task_io_get_inblock(p) +
1168 sig->inblock + sig->cinblock;
1170 task_io_get_oublock(p) +
1171 sig->oublock + sig->coublock;
1172 maxrss = max(sig->maxrss, sig->cmaxrss);
1173 if (psig->cmaxrss < maxrss)
1174 psig->cmaxrss = maxrss;
1175 task_io_accounting_add(&psig->ioac, &p->ioac);
1176 task_io_accounting_add(&psig->ioac, &sig->ioac);
1177 write_sequnlock(&psig->stats_lock);
1178 spin_unlock_irq(¤t->sighand->siglock);
1181 retval = wo->wo_rusage
1182 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1183 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1184 ? p->signal->group_exit_code : p->exit_code;
1185 if (!retval && wo->wo_stat)
1186 retval = put_user(status, wo->wo_stat);
1188 infop = wo->wo_info;
1189 if (!retval && infop)
1190 retval = put_user(SIGCHLD, &infop->si_signo);
1191 if (!retval && infop)
1192 retval = put_user(0, &infop->si_errno);
1193 if (!retval && infop) {
1196 if ((status & 0x7f) == 0) {
1200 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1203 retval = put_user((short)why, &infop->si_code);
1205 retval = put_user(status, &infop->si_status);
1207 if (!retval && infop)
1208 retval = put_user(pid, &infop->si_pid);
1209 if (!retval && infop)
1210 retval = put_user(uid, &infop->si_uid);
1214 if (state == EXIT_TRACE) {
1215 write_lock_irq(&tasklist_lock);
1216 /* We dropped tasklist, ptracer could die and untrace */
1219 /* If parent wants a zombie, don't release it now */
1220 state = EXIT_ZOMBIE;
1221 if (do_notify_parent(p, p->exit_signal))
1223 p->exit_state = state;
1224 write_unlock_irq(&tasklist_lock);
1226 if (state == EXIT_DEAD)
1232 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1235 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1236 return &p->exit_code;
1238 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1239 return &p->signal->group_exit_code;
1245 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1247 * @ptrace: is the wait for ptrace
1248 * @p: task to wait for
1250 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1253 * read_lock(&tasklist_lock), which is released if return value is
1254 * non-zero. Also, grabs and releases @p->sighand->siglock.
1257 * 0 if wait condition didn't exist and search for other wait conditions
1258 * should continue. Non-zero return, -errno on failure and @p's pid on
1259 * success, implies that tasklist_lock is released and wait condition
1260 * search should terminate.
1262 static int wait_task_stopped(struct wait_opts *wo,
1263 int ptrace, struct task_struct *p)
1265 struct siginfo __user *infop;
1266 int retval, exit_code, *p_code, why;
1267 uid_t uid = 0; /* unneeded, required by compiler */
1271 * Traditionally we see ptrace'd stopped tasks regardless of options.
1273 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1276 if (!task_stopped_code(p, ptrace))
1280 spin_lock_irq(&p->sighand->siglock);
1282 p_code = task_stopped_code(p, ptrace);
1283 if (unlikely(!p_code))
1286 exit_code = *p_code;
1290 if (!unlikely(wo->wo_flags & WNOWAIT))
1293 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1295 spin_unlock_irq(&p->sighand->siglock);
1300 * Now we are pretty sure this task is interesting.
1301 * Make sure it doesn't get reaped out from under us while we
1302 * give up the lock and then examine it below. We don't want to
1303 * keep holding onto the tasklist_lock while we call getrusage and
1304 * possibly take page faults for user memory.
1307 pid = task_pid_vnr(p);
1308 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1309 read_unlock(&tasklist_lock);
1310 sched_annotate_sleep();
1312 if (unlikely(wo->wo_flags & WNOWAIT))
1313 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1315 retval = wo->wo_rusage
1316 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1317 if (!retval && wo->wo_stat)
1318 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1320 infop = wo->wo_info;
1321 if (!retval && infop)
1322 retval = put_user(SIGCHLD, &infop->si_signo);
1323 if (!retval && infop)
1324 retval = put_user(0, &infop->si_errno);
1325 if (!retval && infop)
1326 retval = put_user((short)why, &infop->si_code);
1327 if (!retval && infop)
1328 retval = put_user(exit_code, &infop->si_status);
1329 if (!retval && infop)
1330 retval = put_user(pid, &infop->si_pid);
1331 if (!retval && infop)
1332 retval = put_user(uid, &infop->si_uid);
1342 * Handle do_wait work for one task in a live, non-stopped state.
1343 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1344 * the lock and this task is uninteresting. If we return nonzero, we have
1345 * released the lock and the system call should return.
1347 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1353 if (!unlikely(wo->wo_flags & WCONTINUED))
1356 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1359 spin_lock_irq(&p->sighand->siglock);
1360 /* Re-check with the lock held. */
1361 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1362 spin_unlock_irq(&p->sighand->siglock);
1365 if (!unlikely(wo->wo_flags & WNOWAIT))
1366 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1367 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1368 spin_unlock_irq(&p->sighand->siglock);
1370 pid = task_pid_vnr(p);
1372 read_unlock(&tasklist_lock);
1373 sched_annotate_sleep();
1376 retval = wo->wo_rusage
1377 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1379 if (!retval && wo->wo_stat)
1380 retval = put_user(0xffff, wo->wo_stat);
1384 retval = wait_noreap_copyout(wo, p, pid, uid,
1385 CLD_CONTINUED, SIGCONT);
1386 BUG_ON(retval == 0);
1393 * Consider @p for a wait by @parent.
1395 * -ECHILD should be in ->notask_error before the first call.
1396 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1397 * Returns zero if the search for a child should continue;
1398 * then ->notask_error is 0 if @p is an eligible child,
1401 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1402 struct task_struct *p)
1405 * We can race with wait_task_zombie() from another thread.
1406 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1407 * can't confuse the checks below.
1409 int exit_state = ACCESS_ONCE(p->exit_state);
1412 if (unlikely(exit_state == EXIT_DEAD))
1415 ret = eligible_child(wo, ptrace, p);
1419 if (unlikely(exit_state == EXIT_TRACE)) {
1421 * ptrace == 0 means we are the natural parent. In this case
1422 * we should clear notask_error, debugger will notify us.
1424 if (likely(!ptrace))
1425 wo->notask_error = 0;
1429 if (likely(!ptrace) && unlikely(p->ptrace)) {
1431 * If it is traced by its real parent's group, just pretend
1432 * the caller is ptrace_do_wait() and reap this child if it
1435 * This also hides group stop state from real parent; otherwise
1436 * a single stop can be reported twice as group and ptrace stop.
1437 * If a ptracer wants to distinguish these two events for its
1438 * own children it should create a separate process which takes
1439 * the role of real parent.
1441 if (!ptrace_reparented(p))
1446 if (exit_state == EXIT_ZOMBIE) {
1447 /* we don't reap group leaders with subthreads */
1448 if (!delay_group_leader(p)) {
1450 * A zombie ptracee is only visible to its ptracer.
1451 * Notification and reaping will be cascaded to the
1452 * real parent when the ptracer detaches.
1454 if (unlikely(ptrace) || likely(!p->ptrace))
1455 return wait_task_zombie(wo, p);
1459 * Allow access to stopped/continued state via zombie by
1460 * falling through. Clearing of notask_error is complex.
1464 * If WEXITED is set, notask_error should naturally be
1465 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1466 * so, if there are live subthreads, there are events to
1467 * wait for. If all subthreads are dead, it's still safe
1468 * to clear - this function will be called again in finite
1469 * amount time once all the subthreads are released and
1470 * will then return without clearing.
1474 * Stopped state is per-task and thus can't change once the
1475 * target task dies. Only continued and exited can happen.
1476 * Clear notask_error if WCONTINUED | WEXITED.
1478 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1479 wo->notask_error = 0;
1482 * @p is alive and it's gonna stop, continue or exit, so
1483 * there always is something to wait for.
1485 wo->notask_error = 0;
1489 * Wait for stopped. Depending on @ptrace, different stopped state
1490 * is used and the two don't interact with each other.
1492 ret = wait_task_stopped(wo, ptrace, p);
1497 * Wait for continued. There's only one continued state and the
1498 * ptracer can consume it which can confuse the real parent. Don't
1499 * use WCONTINUED from ptracer. You don't need or want it.
1501 return wait_task_continued(wo, p);
1505 * Do the work of do_wait() for one thread in the group, @tsk.
1507 * -ECHILD should be in ->notask_error before the first call.
1508 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1509 * Returns zero if the search for a child should continue; then
1510 * ->notask_error is 0 if there were any eligible children,
1513 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1515 struct task_struct *p;
1517 list_for_each_entry(p, &tsk->children, sibling) {
1518 int ret = wait_consider_task(wo, 0, p);
1527 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1529 struct task_struct *p;
1531 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1532 int ret = wait_consider_task(wo, 1, p);
1541 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1542 int sync, void *key)
1544 struct wait_opts *wo = container_of(wait, struct wait_opts,
1546 struct task_struct *p = key;
1548 if (!eligible_pid(wo, p))
1551 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1554 return default_wake_function(wait, mode, sync, key);
1557 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1559 __wake_up_sync_key(&parent->signal->wait_chldexit,
1560 TASK_INTERRUPTIBLE, 1, p);
1563 static long do_wait(struct wait_opts *wo)
1565 struct task_struct *tsk;
1568 trace_sched_process_wait(wo->wo_pid);
1570 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1571 wo->child_wait.private = current;
1572 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1575 * If there is nothing that can match our criteria, just get out.
1576 * We will clear ->notask_error to zero if we see any child that
1577 * might later match our criteria, even if we are not able to reap
1580 wo->notask_error = -ECHILD;
1581 if ((wo->wo_type < PIDTYPE_MAX) &&
1582 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1585 set_current_state(TASK_INTERRUPTIBLE);
1586 read_lock(&tasklist_lock);
1589 retval = do_wait_thread(wo, tsk);
1593 retval = ptrace_do_wait(wo, tsk);
1597 if (wo->wo_flags & __WNOTHREAD)
1599 } while_each_thread(current, tsk);
1600 read_unlock(&tasklist_lock);
1603 retval = wo->notask_error;
1604 if (!retval && !(wo->wo_flags & WNOHANG)) {
1605 retval = -ERESTARTSYS;
1606 if (!signal_pending(current)) {
1612 __set_current_state(TASK_RUNNING);
1613 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1617 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1618 infop, int, options, struct rusage __user *, ru)
1620 struct wait_opts wo;
1621 struct pid *pid = NULL;
1625 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1626 __WNOTHREAD|__WCLONE|__WALL))
1628 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1641 type = PIDTYPE_PGID;
1649 if (type < PIDTYPE_MAX)
1650 pid = find_get_pid(upid);
1654 wo.wo_flags = options;
1664 * For a WNOHANG return, clear out all the fields
1665 * we would set so the user can easily tell the
1669 ret = put_user(0, &infop->si_signo);
1671 ret = put_user(0, &infop->si_errno);
1673 ret = put_user(0, &infop->si_code);
1675 ret = put_user(0, &infop->si_pid);
1677 ret = put_user(0, &infop->si_uid);
1679 ret = put_user(0, &infop->si_status);
1686 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1687 int, options, struct rusage __user *, ru)
1689 struct wait_opts wo;
1690 struct pid *pid = NULL;
1694 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1695 __WNOTHREAD|__WCLONE|__WALL))
1700 else if (upid < 0) {
1701 type = PIDTYPE_PGID;
1702 pid = find_get_pid(-upid);
1703 } else if (upid == 0) {
1704 type = PIDTYPE_PGID;
1705 pid = get_task_pid(current, PIDTYPE_PGID);
1706 } else /* upid > 0 */ {
1708 pid = find_get_pid(upid);
1713 wo.wo_flags = options | WEXITED;
1715 wo.wo_stat = stat_addr;
1723 #ifdef __ARCH_WANT_SYS_WAITPID
1726 * sys_waitpid() remains for compatibility. waitpid() should be
1727 * implemented by calling sys_wait4() from libc.a.
1729 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1731 return sys_wait4(pid, stat_addr, options, NULL);