4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/iocontext.h>
16 #include <linux/key.h>
17 #include <linux/security.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/init_task.h>
50 #include <linux/perf_event.h>
51 #include <trace/events/sched.h>
52 #include <linux/hw_breakpoint.h>
53 #include <linux/oom.h>
54 #include <linux/writeback.h>
55 #include <linux/shm.h>
56 #include <linux/kcov.h>
57 #include <linux/random.h>
59 #include <asm/uaccess.h>
60 #include <asm/unistd.h>
61 #include <asm/pgtable.h>
62 #include <asm/mmu_context.h>
64 static void __unhash_process(struct task_struct
*p
, bool group_dead
)
67 detach_pid(p
, PIDTYPE_PID
);
69 detach_pid(p
, PIDTYPE_PGID
);
70 detach_pid(p
, PIDTYPE_SID
);
72 list_del_rcu(&p
->tasks
);
73 list_del_init(&p
->sibling
);
74 __this_cpu_dec(process_counts
);
76 list_del_rcu(&p
->thread_group
);
77 list_del_rcu(&p
->thread_node
);
81 * This function expects the tasklist_lock write-locked.
83 static void __exit_signal(struct task_struct
*tsk
)
85 struct signal_struct
*sig
= tsk
->signal
;
86 bool group_dead
= thread_group_leader(tsk
);
87 struct sighand_struct
*sighand
;
88 struct tty_struct
*uninitialized_var(tty
);
89 cputime_t utime
, stime
;
91 sighand
= rcu_dereference_check(tsk
->sighand
,
92 lockdep_tasklist_lock_is_held());
93 spin_lock(&sighand
->siglock
);
95 #ifdef CONFIG_POSIX_TIMERS
96 posix_cpu_timers_exit(tsk
);
98 posix_cpu_timers_exit_group(tsk
);
101 * This can only happen if the caller is de_thread().
102 * FIXME: this is the temporary hack, we should teach
103 * posix-cpu-timers to handle this case correctly.
105 if (unlikely(has_group_leader_pid(tsk
)))
106 posix_cpu_timers_exit_group(tsk
);
115 * If there is any task waiting for the group exit
118 if (sig
->notify_count
> 0 && !--sig
->notify_count
)
119 wake_up_process(sig
->group_exit_task
);
121 if (tsk
== sig
->curr_target
)
122 sig
->curr_target
= next_thread(tsk
);
125 add_device_randomness((const void*) &tsk
->se
.sum_exec_runtime
,
126 sizeof(unsigned long long));
129 * Accumulate here the counters for all threads as they die. We could
130 * skip the group leader because it is the last user of signal_struct,
131 * but we want to avoid the race with thread_group_cputime() which can
132 * see the empty ->thread_head list.
134 task_cputime(tsk
, &utime
, &stime
);
135 write_seqlock(&sig
->stats_lock
);
138 sig
->gtime
+= task_gtime(tsk
);
139 sig
->min_flt
+= tsk
->min_flt
;
140 sig
->maj_flt
+= tsk
->maj_flt
;
141 sig
->nvcsw
+= tsk
->nvcsw
;
142 sig
->nivcsw
+= tsk
->nivcsw
;
143 sig
->inblock
+= task_io_get_inblock(tsk
);
144 sig
->oublock
+= task_io_get_oublock(tsk
);
145 task_io_accounting_add(&sig
->ioac
, &tsk
->ioac
);
146 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
148 __unhash_process(tsk
, group_dead
);
149 write_sequnlock(&sig
->stats_lock
);
152 * Do this under ->siglock, we can race with another thread
153 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
155 flush_sigqueue(&tsk
->pending
);
157 spin_unlock(&sighand
->siglock
);
159 __cleanup_sighand(sighand
);
160 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
162 flush_sigqueue(&sig
->shared_pending
);
167 static void delayed_put_task_struct(struct rcu_head
*rhp
)
169 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
171 perf_event_delayed_put(tsk
);
172 trace_sched_process_free(tsk
);
173 put_task_struct(tsk
);
177 void release_task(struct task_struct
*p
)
179 struct task_struct
*leader
;
182 /* don't need to get the RCU readlock here - the process is dead and
183 * can't be modifying its own credentials. But shut RCU-lockdep up */
185 atomic_dec(&__task_cred(p
)->user
->processes
);
190 write_lock_irq(&tasklist_lock
);
191 ptrace_release_task(p
);
195 * If we are the last non-leader member of the thread
196 * group, and the leader is zombie, then notify the
197 * group leader's parent process. (if it wants notification.)
200 leader
= p
->group_leader
;
201 if (leader
!= p
&& thread_group_empty(leader
)
202 && leader
->exit_state
== EXIT_ZOMBIE
) {
204 * If we were the last child thread and the leader has
205 * exited already, and the leader's parent ignores SIGCHLD,
206 * then we are the one who should release the leader.
208 zap_leader
= do_notify_parent(leader
, leader
->exit_signal
);
210 leader
->exit_state
= EXIT_DEAD
;
213 write_unlock_irq(&tasklist_lock
);
215 call_rcu(&p
->rcu
, delayed_put_task_struct
);
218 if (unlikely(zap_leader
))
223 * Note that if this function returns a valid task_struct pointer (!NULL)
224 * task->usage must remain >0 for the duration of the RCU critical section.
226 struct task_struct
*task_rcu_dereference(struct task_struct
**ptask
)
228 struct sighand_struct
*sighand
;
229 struct task_struct
*task
;
232 * We need to verify that release_task() was not called and thus
233 * delayed_put_task_struct() can't run and drop the last reference
234 * before rcu_read_unlock(). We check task->sighand != NULL,
235 * but we can read the already freed and reused memory.
238 task
= rcu_dereference(*ptask
);
242 probe_kernel_address(&task
->sighand
, sighand
);
245 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
246 * was already freed we can not miss the preceding update of this
250 if (unlikely(task
!= READ_ONCE(*ptask
)))
254 * We've re-checked that "task == *ptask", now we have two different
257 * 1. This is actually the same task/task_struct. In this case
258 * sighand != NULL tells us it is still alive.
260 * 2. This is another task which got the same memory for task_struct.
261 * We can't know this of course, and we can not trust
264 * In this case we actually return a random value, but this is
267 * If we return NULL - we can pretend that we actually noticed that
268 * *ptask was updated when the previous task has exited. Or pretend
269 * that probe_slab_address(&sighand) reads NULL.
271 * If we return the new task (because sighand is not NULL for any
272 * reason) - this is fine too. This (new) task can't go away before
275 * And note: We could even eliminate the false positive if re-read
276 * task->sighand once again to avoid the falsely NULL. But this case
277 * is very unlikely so we don't care.
285 struct task_struct
*try_get_task_struct(struct task_struct
**ptask
)
287 struct task_struct
*task
;
290 task
= task_rcu_dereference(ptask
);
292 get_task_struct(task
);
299 * Determine if a process group is "orphaned", according to the POSIX
300 * definition in 2.2.2.52. Orphaned process groups are not to be affected
301 * by terminal-generated stop signals. Newly orphaned process groups are
302 * to receive a SIGHUP and a SIGCONT.
304 * "I ask you, have you ever known what it is to be an orphan?"
306 static int will_become_orphaned_pgrp(struct pid
*pgrp
,
307 struct task_struct
*ignored_task
)
309 struct task_struct
*p
;
311 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
312 if ((p
== ignored_task
) ||
313 (p
->exit_state
&& thread_group_empty(p
)) ||
314 is_global_init(p
->real_parent
))
317 if (task_pgrp(p
->real_parent
) != pgrp
&&
318 task_session(p
->real_parent
) == task_session(p
))
320 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
325 int is_current_pgrp_orphaned(void)
329 read_lock(&tasklist_lock
);
330 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
331 read_unlock(&tasklist_lock
);
336 static bool has_stopped_jobs(struct pid
*pgrp
)
338 struct task_struct
*p
;
340 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
341 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
343 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
349 * Check to see if any process groups have become orphaned as
350 * a result of our exiting, and if they have any stopped jobs,
351 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
354 kill_orphaned_pgrp(struct task_struct
*tsk
, struct task_struct
*parent
)
356 struct pid
*pgrp
= task_pgrp(tsk
);
357 struct task_struct
*ignored_task
= tsk
;
360 /* exit: our father is in a different pgrp than
361 * we are and we were the only connection outside.
363 parent
= tsk
->real_parent
;
365 /* reparent: our child is in a different pgrp than
366 * we are, and it was the only connection outside.
370 if (task_pgrp(parent
) != pgrp
&&
371 task_session(parent
) == task_session(tsk
) &&
372 will_become_orphaned_pgrp(pgrp
, ignored_task
) &&
373 has_stopped_jobs(pgrp
)) {
374 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
375 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
381 * A task is exiting. If it owned this mm, find a new owner for the mm.
383 void mm_update_next_owner(struct mm_struct
*mm
)
385 struct task_struct
*c
, *g
, *p
= current
;
389 * If the exiting or execing task is not the owner, it's
390 * someone else's problem.
395 * The current owner is exiting/execing and there are no other
396 * candidates. Do not leave the mm pointing to a possibly
397 * freed task structure.
399 if (atomic_read(&mm
->mm_users
) <= 1) {
404 read_lock(&tasklist_lock
);
406 * Search in the children
408 list_for_each_entry(c
, &p
->children
, sibling
) {
410 goto assign_new_owner
;
414 * Search in the siblings
416 list_for_each_entry(c
, &p
->real_parent
->children
, sibling
) {
418 goto assign_new_owner
;
422 * Search through everything else, we should not get here often.
424 for_each_process(g
) {
425 if (g
->flags
& PF_KTHREAD
)
427 for_each_thread(g
, c
) {
429 goto assign_new_owner
;
434 read_unlock(&tasklist_lock
);
436 * We found no owner yet mm_users > 1: this implies that we are
437 * most likely racing with swapoff (try_to_unuse()) or /proc or
438 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
447 * The task_lock protects c->mm from changing.
448 * We always want mm->owner->mm == mm
452 * Delay read_unlock() till we have the task_lock()
453 * to ensure that c does not slip away underneath us
455 read_unlock(&tasklist_lock
);
465 #endif /* CONFIG_MEMCG */
468 * Turn us into a lazy TLB process if we
471 static void exit_mm(struct task_struct
*tsk
)
473 struct mm_struct
*mm
= tsk
->mm
;
474 struct core_state
*core_state
;
481 * Serialize with any possible pending coredump.
482 * We must hold mmap_sem around checking core_state
483 * and clearing tsk->mm. The core-inducing thread
484 * will increment ->nr_threads for each thread in the
485 * group with ->mm != NULL.
487 down_read(&mm
->mmap_sem
);
488 core_state
= mm
->core_state
;
490 struct core_thread self
;
492 up_read(&mm
->mmap_sem
);
495 self
.next
= xchg(&core_state
->dumper
.next
, &self
);
497 * Implies mb(), the result of xchg() must be visible
498 * to core_state->dumper.
500 if (atomic_dec_and_test(&core_state
->nr_threads
))
501 complete(&core_state
->startup
);
504 set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
505 if (!self
.task
) /* see coredump_finish() */
507 freezable_schedule();
509 __set_task_state(tsk
, TASK_RUNNING
);
510 down_read(&mm
->mmap_sem
);
512 atomic_inc(&mm
->mm_count
);
513 BUG_ON(mm
!= tsk
->active_mm
);
514 /* more a memory barrier than a real lock */
517 up_read(&mm
->mmap_sem
);
518 enter_lazy_tlb(mm
, current
);
520 mm_update_next_owner(mm
);
522 if (test_thread_flag(TIF_MEMDIE
))
526 static struct task_struct
*find_alive_thread(struct task_struct
*p
)
528 struct task_struct
*t
;
530 for_each_thread(p
, t
) {
531 if (!(t
->flags
& PF_EXITING
))
537 static struct task_struct
*find_child_reaper(struct task_struct
*father
)
538 __releases(&tasklist_lock
)
539 __acquires(&tasklist_lock
)
541 struct pid_namespace
*pid_ns
= task_active_pid_ns(father
);
542 struct task_struct
*reaper
= pid_ns
->child_reaper
;
544 if (likely(reaper
!= father
))
547 reaper
= find_alive_thread(father
);
549 pid_ns
->child_reaper
= reaper
;
553 write_unlock_irq(&tasklist_lock
);
554 if (unlikely(pid_ns
== &init_pid_ns
)) {
555 panic("Attempted to kill init! exitcode=0x%08x\n",
556 father
->signal
->group_exit_code
?: father
->exit_code
);
558 zap_pid_ns_processes(pid_ns
);
559 write_lock_irq(&tasklist_lock
);
565 * When we die, we re-parent all our children, and try to:
566 * 1. give them to another thread in our thread group, if such a member exists
567 * 2. give it to the first ancestor process which prctl'd itself as a
568 * child_subreaper for its children (like a service manager)
569 * 3. give it to the init process (PID 1) in our pid namespace
571 static struct task_struct
*find_new_reaper(struct task_struct
*father
,
572 struct task_struct
*child_reaper
)
574 struct task_struct
*thread
, *reaper
;
576 thread
= find_alive_thread(father
);
580 if (father
->signal
->has_child_subreaper
) {
582 * Find the first ->is_child_subreaper ancestor in our pid_ns.
583 * We start from father to ensure we can not look into another
584 * namespace, this is safe because all its threads are dead.
586 for (reaper
= father
;
587 !same_thread_group(reaper
, child_reaper
);
588 reaper
= reaper
->real_parent
) {
589 /* call_usermodehelper() descendants need this check */
590 if (reaper
== &init_task
)
592 if (!reaper
->signal
->is_child_subreaper
)
594 thread
= find_alive_thread(reaper
);
604 * Any that need to be release_task'd are put on the @dead list.
606 static void reparent_leader(struct task_struct
*father
, struct task_struct
*p
,
607 struct list_head
*dead
)
609 if (unlikely(p
->exit_state
== EXIT_DEAD
))
612 /* We don't want people slaying init. */
613 p
->exit_signal
= SIGCHLD
;
615 /* If it has exited notify the new parent about this child's death. */
617 p
->exit_state
== EXIT_ZOMBIE
&& thread_group_empty(p
)) {
618 if (do_notify_parent(p
, p
->exit_signal
)) {
619 p
->exit_state
= EXIT_DEAD
;
620 list_add(&p
->ptrace_entry
, dead
);
624 kill_orphaned_pgrp(p
, father
);
628 * This does two things:
630 * A. Make init inherit all the child processes
631 * B. Check to see if any process groups have become orphaned
632 * as a result of our exiting, and if they have any stopped
633 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
635 static void forget_original_parent(struct task_struct
*father
,
636 struct list_head
*dead
)
638 struct task_struct
*p
, *t
, *reaper
;
640 if (unlikely(!list_empty(&father
->ptraced
)))
641 exit_ptrace(father
, dead
);
643 /* Can drop and reacquire tasklist_lock */
644 reaper
= find_child_reaper(father
);
645 if (list_empty(&father
->children
))
648 reaper
= find_new_reaper(father
, reaper
);
649 list_for_each_entry(p
, &father
->children
, sibling
) {
650 for_each_thread(p
, t
) {
651 t
->real_parent
= reaper
;
652 BUG_ON((!t
->ptrace
) != (t
->parent
== father
));
653 if (likely(!t
->ptrace
))
654 t
->parent
= t
->real_parent
;
655 if (t
->pdeath_signal
)
656 group_send_sig_info(t
->pdeath_signal
,
660 * If this is a threaded reparent there is no need to
661 * notify anyone anything has happened.
663 if (!same_thread_group(reaper
, father
))
664 reparent_leader(father
, p
, dead
);
666 list_splice_tail_init(&father
->children
, &reaper
->children
);
670 * Send signals to all our closest relatives so that they know
671 * to properly mourn us..
673 static void exit_notify(struct task_struct
*tsk
, int group_dead
)
676 struct task_struct
*p
, *n
;
679 write_lock_irq(&tasklist_lock
);
680 forget_original_parent(tsk
, &dead
);
683 kill_orphaned_pgrp(tsk
->group_leader
, NULL
);
685 if (unlikely(tsk
->ptrace
)) {
686 int sig
= thread_group_leader(tsk
) &&
687 thread_group_empty(tsk
) &&
688 !ptrace_reparented(tsk
) ?
689 tsk
->exit_signal
: SIGCHLD
;
690 autoreap
= do_notify_parent(tsk
, sig
);
691 } else if (thread_group_leader(tsk
)) {
692 autoreap
= thread_group_empty(tsk
) &&
693 do_notify_parent(tsk
, tsk
->exit_signal
);
698 tsk
->exit_state
= autoreap
? EXIT_DEAD
: EXIT_ZOMBIE
;
699 if (tsk
->exit_state
== EXIT_DEAD
)
700 list_add(&tsk
->ptrace_entry
, &dead
);
702 /* mt-exec, de_thread() is waiting for group leader */
703 if (unlikely(tsk
->signal
->notify_count
< 0))
704 wake_up_process(tsk
->signal
->group_exit_task
);
705 write_unlock_irq(&tasklist_lock
);
707 list_for_each_entry_safe(p
, n
, &dead
, ptrace_entry
) {
708 list_del_init(&p
->ptrace_entry
);
713 #ifdef CONFIG_DEBUG_STACK_USAGE
714 static void check_stack_usage(void)
716 static DEFINE_SPINLOCK(low_water_lock
);
717 static int lowest_to_date
= THREAD_SIZE
;
720 free
= stack_not_used(current
);
722 if (free
>= lowest_to_date
)
725 spin_lock(&low_water_lock
);
726 if (free
< lowest_to_date
) {
727 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
728 current
->comm
, task_pid_nr(current
), free
);
729 lowest_to_date
= free
;
731 spin_unlock(&low_water_lock
);
734 static inline void check_stack_usage(void) {}
737 void __noreturn
do_exit(long code
)
739 struct task_struct
*tsk
= current
;
741 TASKS_RCU(int tasks_rcu_i
);
743 profile_task_exit(tsk
);
746 WARN_ON(blk_needs_flush_plug(tsk
));
748 if (unlikely(in_interrupt()))
749 panic("Aiee, killing interrupt handler!");
750 if (unlikely(!tsk
->pid
))
751 panic("Attempted to kill the idle task!");
754 * If do_exit is called because this processes oopsed, it's possible
755 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
756 * continuing. Amongst other possible reasons, this is to prevent
757 * mm_release()->clear_child_tid() from writing to a user-controlled
762 ptrace_event(PTRACE_EVENT_EXIT
, code
);
764 validate_creds_for_do_exit(tsk
);
767 * We're taking recursive faults here in do_exit. Safest is to just
768 * leave this task alone and wait for reboot.
770 if (unlikely(tsk
->flags
& PF_EXITING
)) {
771 pr_alert("Fixing recursive fault but reboot is needed!\n");
773 * We can do this unlocked here. The futex code uses
774 * this flag just to verify whether the pi state
775 * cleanup has been done or not. In the worst case it
776 * loops once more. We pretend that the cleanup was
777 * done as there is no way to return. Either the
778 * OWNER_DIED bit is set by now or we push the blocked
779 * task into the wait for ever nirwana as well.
781 tsk
->flags
|= PF_EXITPIDONE
;
782 set_current_state(TASK_UNINTERRUPTIBLE
);
786 exit_signals(tsk
); /* sets PF_EXITING */
788 * Ensure that all new tsk->pi_lock acquisitions must observe
789 * PF_EXITING. Serializes against futex.c:attach_to_pi_owner().
793 * Ensure that we must observe the pi_state in exit_mm() ->
794 * mm_release() -> exit_pi_state_list().
796 raw_spin_unlock_wait(&tsk
->pi_lock
);
798 if (unlikely(in_atomic())) {
799 pr_info("note: %s[%d] exited with preempt_count %d\n",
800 current
->comm
, task_pid_nr(current
),
802 preempt_count_set(PREEMPT_ENABLED
);
805 /* sync mm's RSS info before statistics gathering */
807 sync_mm_rss(tsk
->mm
);
808 acct_update_integrals(tsk
);
809 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
811 #ifdef CONFIG_POSIX_TIMERS
812 hrtimer_cancel(&tsk
->signal
->real_timer
);
813 exit_itimers(tsk
->signal
);
816 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, tsk
->mm
);
818 acct_collect(code
, group_dead
);
823 tsk
->exit_code
= code
;
824 taskstats_exit(tsk
, group_dead
);
830 trace_sched_process_exit(tsk
);
837 disassociate_ctty(1);
838 exit_task_namespaces(tsk
);
843 * Flush inherited counters to the parent - before the parent
844 * gets woken up by child-exit notifications.
846 * because of cgroup mode, must be called before cgroup_exit()
848 perf_event_exit_task(tsk
);
850 sched_autogroup_exit_task(tsk
);
854 * FIXME: do that only when needed, using sched_exit tracepoint
856 flush_ptrace_hw_breakpoint(tsk
);
858 TASKS_RCU(preempt_disable());
859 TASKS_RCU(tasks_rcu_i
= __srcu_read_lock(&tasks_rcu_exit_srcu
));
860 TASKS_RCU(preempt_enable());
861 exit_notify(tsk
, group_dead
);
862 proc_exit_connector(tsk
);
863 mpol_put_task_policy(tsk
);
865 if (unlikely(current
->pi_state_cache
))
866 kfree(current
->pi_state_cache
);
869 * Make sure we are holding no locks:
871 debug_check_no_locks_held();
873 * We can do this unlocked here. The futex code uses this flag
874 * just to verify whether the pi state cleanup has been done
875 * or not. In the worst case it loops once more.
877 tsk
->flags
|= PF_EXITPIDONE
;
880 exit_io_context(tsk
);
882 if (tsk
->splice_pipe
)
883 free_pipe_info(tsk
->splice_pipe
);
885 if (tsk
->task_frag
.page
)
886 put_page(tsk
->task_frag
.page
);
888 validate_creds_for_do_exit(tsk
);
893 __this_cpu_add(dirty_throttle_leaks
, tsk
->nr_dirtied
);
895 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu
, tasks_rcu_i
));
899 EXPORT_SYMBOL_GPL(do_exit
);
901 void complete_and_exit(struct completion
*comp
, long code
)
908 EXPORT_SYMBOL(complete_and_exit
);
910 SYSCALL_DEFINE1(exit
, int, error_code
)
912 do_exit((error_code
&0xff)<<8);
916 * Take down every thread in the group. This is called by fatal signals
917 * as well as by sys_exit_group (below).
920 do_group_exit(int exit_code
)
922 struct signal_struct
*sig
= current
->signal
;
924 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
926 if (signal_group_exit(sig
))
927 exit_code
= sig
->group_exit_code
;
928 else if (!thread_group_empty(current
)) {
929 struct sighand_struct
*const sighand
= current
->sighand
;
931 spin_lock_irq(&sighand
->siglock
);
932 if (signal_group_exit(sig
))
933 /* Another thread got here before we took the lock. */
934 exit_code
= sig
->group_exit_code
;
936 sig
->group_exit_code
= exit_code
;
937 sig
->flags
= SIGNAL_GROUP_EXIT
;
938 zap_other_threads(current
);
940 spin_unlock_irq(&sighand
->siglock
);
948 * this kills every thread in the thread group. Note that any externally
949 * wait4()-ing process will get the correct exit code - even if this
950 * thread is not the thread group leader.
952 SYSCALL_DEFINE1(exit_group
, int, error_code
)
954 do_group_exit((error_code
& 0xff) << 8);
960 enum pid_type wo_type
;
964 struct siginfo __user
*wo_info
;
966 struct rusage __user
*wo_rusage
;
968 wait_queue_t child_wait
;
973 struct pid
*task_pid_type(struct task_struct
*task
, enum pid_type type
)
975 if (type
!= PIDTYPE_PID
)
976 task
= task
->group_leader
;
977 return task
->pids
[type
].pid
;
980 static int eligible_pid(struct wait_opts
*wo
, struct task_struct
*p
)
982 return wo
->wo_type
== PIDTYPE_MAX
||
983 task_pid_type(p
, wo
->wo_type
) == wo
->wo_pid
;
987 eligible_child(struct wait_opts
*wo
, bool ptrace
, struct task_struct
*p
)
989 if (!eligible_pid(wo
, p
))
993 * Wait for all children (clone and not) if __WALL is set or
994 * if it is traced by us.
996 if (ptrace
|| (wo
->wo_flags
& __WALL
))
1000 * Otherwise, wait for clone children *only* if __WCLONE is set;
1001 * otherwise, wait for non-clone children *only*.
1003 * Note: a "clone" child here is one that reports to its parent
1004 * using a signal other than SIGCHLD, or a non-leader thread which
1005 * we can only see if it is traced by us.
1007 if ((p
->exit_signal
!= SIGCHLD
) ^ !!(wo
->wo_flags
& __WCLONE
))
1013 static int wait_noreap_copyout(struct wait_opts
*wo
, struct task_struct
*p
,
1014 pid_t pid
, uid_t uid
, int why
, int status
)
1016 struct siginfo __user
*infop
;
1017 int retval
= wo
->wo_rusage
1018 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1021 infop
= wo
->wo_info
;
1024 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1026 retval
= put_user(0, &infop
->si_errno
);
1028 retval
= put_user((short)why
, &infop
->si_code
);
1030 retval
= put_user(pid
, &infop
->si_pid
);
1032 retval
= put_user(uid
, &infop
->si_uid
);
1034 retval
= put_user(status
, &infop
->si_status
);
1042 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1043 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1044 * the lock and this task is uninteresting. If we return nonzero, we have
1045 * released the lock and the system call should return.
1047 static int wait_task_zombie(struct wait_opts
*wo
, struct task_struct
*p
)
1049 int state
, retval
, status
;
1050 pid_t pid
= task_pid_vnr(p
);
1051 uid_t uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1052 struct siginfo __user
*infop
;
1054 if (!likely(wo
->wo_flags
& WEXITED
))
1057 if (unlikely(wo
->wo_flags
& WNOWAIT
)) {
1058 int exit_code
= p
->exit_code
;
1062 read_unlock(&tasklist_lock
);
1063 sched_annotate_sleep();
1065 if ((exit_code
& 0x7f) == 0) {
1067 status
= exit_code
>> 8;
1069 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1070 status
= exit_code
& 0x7f;
1072 return wait_noreap_copyout(wo
, p
, pid
, uid
, why
, status
);
1075 * Move the task's state to DEAD/TRACE, only one thread can do this.
1077 state
= (ptrace_reparented(p
) && thread_group_leader(p
)) ?
1078 EXIT_TRACE
: EXIT_DEAD
;
1079 if (cmpxchg(&p
->exit_state
, EXIT_ZOMBIE
, state
) != EXIT_ZOMBIE
)
1082 * We own this thread, nobody else can reap it.
1084 read_unlock(&tasklist_lock
);
1085 sched_annotate_sleep();
1088 * Check thread_group_leader() to exclude the traced sub-threads.
1090 if (state
== EXIT_DEAD
&& thread_group_leader(p
)) {
1091 struct signal_struct
*sig
= p
->signal
;
1092 struct signal_struct
*psig
= current
->signal
;
1093 unsigned long maxrss
;
1094 cputime_t tgutime
, tgstime
;
1097 * The resource counters for the group leader are in its
1098 * own task_struct. Those for dead threads in the group
1099 * are in its signal_struct, as are those for the child
1100 * processes it has previously reaped. All these
1101 * accumulate in the parent's signal_struct c* fields.
1103 * We don't bother to take a lock here to protect these
1104 * p->signal fields because the whole thread group is dead
1105 * and nobody can change them.
1107 * psig->stats_lock also protects us from our sub-theads
1108 * which can reap other children at the same time. Until
1109 * we change k_getrusage()-like users to rely on this lock
1110 * we have to take ->siglock as well.
1112 * We use thread_group_cputime_adjusted() to get times for
1113 * the thread group, which consolidates times for all threads
1114 * in the group including the group leader.
1116 thread_group_cputime_adjusted(p
, &tgutime
, &tgstime
);
1117 spin_lock_irq(¤t
->sighand
->siglock
);
1118 write_seqlock(&psig
->stats_lock
);
1119 psig
->cutime
+= tgutime
+ sig
->cutime
;
1120 psig
->cstime
+= tgstime
+ sig
->cstime
;
1121 psig
->cgtime
+= task_gtime(p
) + sig
->gtime
+ sig
->cgtime
;
1123 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1125 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1127 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1129 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1131 task_io_get_inblock(p
) +
1132 sig
->inblock
+ sig
->cinblock
;
1134 task_io_get_oublock(p
) +
1135 sig
->oublock
+ sig
->coublock
;
1136 maxrss
= max(sig
->maxrss
, sig
->cmaxrss
);
1137 if (psig
->cmaxrss
< maxrss
)
1138 psig
->cmaxrss
= maxrss
;
1139 task_io_accounting_add(&psig
->ioac
, &p
->ioac
);
1140 task_io_accounting_add(&psig
->ioac
, &sig
->ioac
);
1141 write_sequnlock(&psig
->stats_lock
);
1142 spin_unlock_irq(¤t
->sighand
->siglock
);
1145 retval
= wo
->wo_rusage
1146 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1147 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1148 ? p
->signal
->group_exit_code
: p
->exit_code
;
1149 if (!retval
&& wo
->wo_stat
)
1150 retval
= put_user(status
, wo
->wo_stat
);
1152 infop
= wo
->wo_info
;
1153 if (!retval
&& infop
)
1154 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1155 if (!retval
&& infop
)
1156 retval
= put_user(0, &infop
->si_errno
);
1157 if (!retval
&& infop
) {
1160 if ((status
& 0x7f) == 0) {
1164 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1167 retval
= put_user((short)why
, &infop
->si_code
);
1169 retval
= put_user(status
, &infop
->si_status
);
1171 if (!retval
&& infop
)
1172 retval
= put_user(pid
, &infop
->si_pid
);
1173 if (!retval
&& infop
)
1174 retval
= put_user(uid
, &infop
->si_uid
);
1178 if (state
== EXIT_TRACE
) {
1179 write_lock_irq(&tasklist_lock
);
1180 /* We dropped tasklist, ptracer could die and untrace */
1183 /* If parent wants a zombie, don't release it now */
1184 state
= EXIT_ZOMBIE
;
1185 if (do_notify_parent(p
, p
->exit_signal
))
1187 p
->exit_state
= state
;
1188 write_unlock_irq(&tasklist_lock
);
1190 if (state
== EXIT_DEAD
)
1196 static int *task_stopped_code(struct task_struct
*p
, bool ptrace
)
1199 if (task_is_traced(p
) && !(p
->jobctl
& JOBCTL_LISTENING
))
1200 return &p
->exit_code
;
1202 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
1203 return &p
->signal
->group_exit_code
;
1209 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1211 * @ptrace: is the wait for ptrace
1212 * @p: task to wait for
1214 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1217 * read_lock(&tasklist_lock), which is released if return value is
1218 * non-zero. Also, grabs and releases @p->sighand->siglock.
1221 * 0 if wait condition didn't exist and search for other wait conditions
1222 * should continue. Non-zero return, -errno on failure and @p's pid on
1223 * success, implies that tasklist_lock is released and wait condition
1224 * search should terminate.
1226 static int wait_task_stopped(struct wait_opts
*wo
,
1227 int ptrace
, struct task_struct
*p
)
1229 struct siginfo __user
*infop
;
1230 int retval
, exit_code
, *p_code
, why
;
1231 uid_t uid
= 0; /* unneeded, required by compiler */
1235 * Traditionally we see ptrace'd stopped tasks regardless of options.
1237 if (!ptrace
&& !(wo
->wo_flags
& WUNTRACED
))
1240 if (!task_stopped_code(p
, ptrace
))
1244 spin_lock_irq(&p
->sighand
->siglock
);
1246 p_code
= task_stopped_code(p
, ptrace
);
1247 if (unlikely(!p_code
))
1250 exit_code
= *p_code
;
1254 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1257 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1259 spin_unlock_irq(&p
->sighand
->siglock
);
1264 * Now we are pretty sure this task is interesting.
1265 * Make sure it doesn't get reaped out from under us while we
1266 * give up the lock and then examine it below. We don't want to
1267 * keep holding onto the tasklist_lock while we call getrusage and
1268 * possibly take page faults for user memory.
1271 pid
= task_pid_vnr(p
);
1272 why
= ptrace
? CLD_TRAPPED
: CLD_STOPPED
;
1273 read_unlock(&tasklist_lock
);
1274 sched_annotate_sleep();
1276 if (unlikely(wo
->wo_flags
& WNOWAIT
))
1277 return wait_noreap_copyout(wo
, p
, pid
, uid
, why
, exit_code
);
1279 retval
= wo
->wo_rusage
1280 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1281 if (!retval
&& wo
->wo_stat
)
1282 retval
= put_user((exit_code
<< 8) | 0x7f, wo
->wo_stat
);
1284 infop
= wo
->wo_info
;
1285 if (!retval
&& infop
)
1286 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1287 if (!retval
&& infop
)
1288 retval
= put_user(0, &infop
->si_errno
);
1289 if (!retval
&& infop
)
1290 retval
= put_user((short)why
, &infop
->si_code
);
1291 if (!retval
&& infop
)
1292 retval
= put_user(exit_code
, &infop
->si_status
);
1293 if (!retval
&& infop
)
1294 retval
= put_user(pid
, &infop
->si_pid
);
1295 if (!retval
&& infop
)
1296 retval
= put_user(uid
, &infop
->si_uid
);
1306 * Handle do_wait work for one task in a live, non-stopped state.
1307 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1308 * the lock and this task is uninteresting. If we return nonzero, we have
1309 * released the lock and the system call should return.
1311 static int wait_task_continued(struct wait_opts
*wo
, struct task_struct
*p
)
1317 if (!unlikely(wo
->wo_flags
& WCONTINUED
))
1320 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1323 spin_lock_irq(&p
->sighand
->siglock
);
1324 /* Re-check with the lock held. */
1325 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1326 spin_unlock_irq(&p
->sighand
->siglock
);
1329 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1330 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1331 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1332 spin_unlock_irq(&p
->sighand
->siglock
);
1334 pid
= task_pid_vnr(p
);
1336 read_unlock(&tasklist_lock
);
1337 sched_annotate_sleep();
1340 retval
= wo
->wo_rusage
1341 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1343 if (!retval
&& wo
->wo_stat
)
1344 retval
= put_user(0xffff, wo
->wo_stat
);
1348 retval
= wait_noreap_copyout(wo
, p
, pid
, uid
,
1349 CLD_CONTINUED
, SIGCONT
);
1350 BUG_ON(retval
== 0);
1357 * Consider @p for a wait by @parent.
1359 * -ECHILD should be in ->notask_error before the first call.
1360 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1361 * Returns zero if the search for a child should continue;
1362 * then ->notask_error is 0 if @p is an eligible child,
1363 * or another error from security_task_wait(), or still -ECHILD.
1365 static int wait_consider_task(struct wait_opts
*wo
, int ptrace
,
1366 struct task_struct
*p
)
1369 * We can race with wait_task_zombie() from another thread.
1370 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1371 * can't confuse the checks below.
1373 int exit_state
= ACCESS_ONCE(p
->exit_state
);
1376 if (unlikely(exit_state
== EXIT_DEAD
))
1379 ret
= eligible_child(wo
, ptrace
, p
);
1383 ret
= security_task_wait(p
);
1384 if (unlikely(ret
< 0)) {
1386 * If we have not yet seen any eligible child,
1387 * then let this error code replace -ECHILD.
1388 * A permission error will give the user a clue
1389 * to look for security policy problems, rather
1390 * than for mysterious wait bugs.
1392 if (wo
->notask_error
)
1393 wo
->notask_error
= ret
;
1397 if (unlikely(exit_state
== EXIT_TRACE
)) {
1399 * ptrace == 0 means we are the natural parent. In this case
1400 * we should clear notask_error, debugger will notify us.
1402 if (likely(!ptrace
))
1403 wo
->notask_error
= 0;
1407 if (likely(!ptrace
) && unlikely(p
->ptrace
)) {
1409 * If it is traced by its real parent's group, just pretend
1410 * the caller is ptrace_do_wait() and reap this child if it
1413 * This also hides group stop state from real parent; otherwise
1414 * a single stop can be reported twice as group and ptrace stop.
1415 * If a ptracer wants to distinguish these two events for its
1416 * own children it should create a separate process which takes
1417 * the role of real parent.
1419 if (!ptrace_reparented(p
))
1424 if (exit_state
== EXIT_ZOMBIE
) {
1425 /* we don't reap group leaders with subthreads */
1426 if (!delay_group_leader(p
)) {
1428 * A zombie ptracee is only visible to its ptracer.
1429 * Notification and reaping will be cascaded to the
1430 * real parent when the ptracer detaches.
1432 if (unlikely(ptrace
) || likely(!p
->ptrace
))
1433 return wait_task_zombie(wo
, p
);
1437 * Allow access to stopped/continued state via zombie by
1438 * falling through. Clearing of notask_error is complex.
1442 * If WEXITED is set, notask_error should naturally be
1443 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1444 * so, if there are live subthreads, there are events to
1445 * wait for. If all subthreads are dead, it's still safe
1446 * to clear - this function will be called again in finite
1447 * amount time once all the subthreads are released and
1448 * will then return without clearing.
1452 * Stopped state is per-task and thus can't change once the
1453 * target task dies. Only continued and exited can happen.
1454 * Clear notask_error if WCONTINUED | WEXITED.
1456 if (likely(!ptrace
) || (wo
->wo_flags
& (WCONTINUED
| WEXITED
)))
1457 wo
->notask_error
= 0;
1460 * @p is alive and it's gonna stop, continue or exit, so
1461 * there always is something to wait for.
1463 wo
->notask_error
= 0;
1467 * Wait for stopped. Depending on @ptrace, different stopped state
1468 * is used and the two don't interact with each other.
1470 ret
= wait_task_stopped(wo
, ptrace
, p
);
1475 * Wait for continued. There's only one continued state and the
1476 * ptracer can consume it which can confuse the real parent. Don't
1477 * use WCONTINUED from ptracer. You don't need or want it.
1479 return wait_task_continued(wo
, p
);
1483 * Do the work of do_wait() for one thread in the group, @tsk.
1485 * -ECHILD should be in ->notask_error before the first call.
1486 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1487 * Returns zero if the search for a child should continue; then
1488 * ->notask_error is 0 if there were any eligible children,
1489 * or another error from security_task_wait(), or still -ECHILD.
1491 static int do_wait_thread(struct wait_opts
*wo
, struct task_struct
*tsk
)
1493 struct task_struct
*p
;
1495 list_for_each_entry(p
, &tsk
->children
, sibling
) {
1496 int ret
= wait_consider_task(wo
, 0, p
);
1505 static int ptrace_do_wait(struct wait_opts
*wo
, struct task_struct
*tsk
)
1507 struct task_struct
*p
;
1509 list_for_each_entry(p
, &tsk
->ptraced
, ptrace_entry
) {
1510 int ret
= wait_consider_task(wo
, 1, p
);
1519 static int child_wait_callback(wait_queue_t
*wait
, unsigned mode
,
1520 int sync
, void *key
)
1522 struct wait_opts
*wo
= container_of(wait
, struct wait_opts
,
1524 struct task_struct
*p
= key
;
1526 if (!eligible_pid(wo
, p
))
1529 if ((wo
->wo_flags
& __WNOTHREAD
) && wait
->private != p
->parent
)
1532 return default_wake_function(wait
, mode
, sync
, key
);
1535 void __wake_up_parent(struct task_struct
*p
, struct task_struct
*parent
)
1537 __wake_up_sync_key(&parent
->signal
->wait_chldexit
,
1538 TASK_INTERRUPTIBLE
, 1, p
);
1541 static long do_wait(struct wait_opts
*wo
)
1543 struct task_struct
*tsk
;
1546 trace_sched_process_wait(wo
->wo_pid
);
1548 init_waitqueue_func_entry(&wo
->child_wait
, child_wait_callback
);
1549 wo
->child_wait
.private = current
;
1550 add_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1553 * If there is nothing that can match our criteria, just get out.
1554 * We will clear ->notask_error to zero if we see any child that
1555 * might later match our criteria, even if we are not able to reap
1558 wo
->notask_error
= -ECHILD
;
1559 if ((wo
->wo_type
< PIDTYPE_MAX
) &&
1560 (!wo
->wo_pid
|| hlist_empty(&wo
->wo_pid
->tasks
[wo
->wo_type
])))
1563 set_current_state(TASK_INTERRUPTIBLE
);
1564 read_lock(&tasklist_lock
);
1567 retval
= do_wait_thread(wo
, tsk
);
1571 retval
= ptrace_do_wait(wo
, tsk
);
1575 if (wo
->wo_flags
& __WNOTHREAD
)
1577 } while_each_thread(current
, tsk
);
1578 read_unlock(&tasklist_lock
);
1581 retval
= wo
->notask_error
;
1582 if (!retval
&& !(wo
->wo_flags
& WNOHANG
)) {
1583 retval
= -ERESTARTSYS
;
1584 if (!signal_pending(current
)) {
1590 __set_current_state(TASK_RUNNING
);
1591 remove_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1595 SYSCALL_DEFINE5(waitid
, int, which
, pid_t
, upid
, struct siginfo __user
*,
1596 infop
, int, options
, struct rusage __user
*, ru
)
1598 struct wait_opts wo
;
1599 struct pid
*pid
= NULL
;
1603 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
|
1604 __WNOTHREAD
|__WCLONE
|__WALL
))
1606 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1619 type
= PIDTYPE_PGID
;
1627 if (type
< PIDTYPE_MAX
)
1628 pid
= find_get_pid(upid
);
1632 wo
.wo_flags
= options
;
1642 * For a WNOHANG return, clear out all the fields
1643 * we would set so the user can easily tell the
1647 ret
= put_user(0, &infop
->si_signo
);
1649 ret
= put_user(0, &infop
->si_errno
);
1651 ret
= put_user(0, &infop
->si_code
);
1653 ret
= put_user(0, &infop
->si_pid
);
1655 ret
= put_user(0, &infop
->si_uid
);
1657 ret
= put_user(0, &infop
->si_status
);
1664 SYSCALL_DEFINE4(wait4
, pid_t
, upid
, int __user
*, stat_addr
,
1665 int, options
, struct rusage __user
*, ru
)
1667 struct wait_opts wo
;
1668 struct pid
*pid
= NULL
;
1672 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1673 __WNOTHREAD
|__WCLONE
|__WALL
))
1678 else if (upid
< 0) {
1679 type
= PIDTYPE_PGID
;
1680 pid
= find_get_pid(-upid
);
1681 } else if (upid
== 0) {
1682 type
= PIDTYPE_PGID
;
1683 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1684 } else /* upid > 0 */ {
1686 pid
= find_get_pid(upid
);
1691 wo
.wo_flags
= options
| WEXITED
;
1693 wo
.wo_stat
= stat_addr
;
1701 #ifdef __ARCH_WANT_SYS_WAITPID
1704 * sys_waitpid() remains for compatibility. waitpid() should be
1705 * implemented by calling sys_wait4() from libc.a.
1707 SYSCALL_DEFINE3(waitpid
, pid_t
, pid
, int __user
*, stat_addr
, int, options
)
1709 return sys_wait4(pid
, stat_addr
, options
, NULL
);