4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/sched/autogroup.h>
10 #include <linux/sched/mm.h>
11 #include <linux/sched/stat.h>
12 #include <linux/sched/task.h>
13 #include <linux/sched/task_stack.h>
14 #include <linux/sched/cputime.h>
15 #include <linux/interrupt.h>
16 #include <linux/module.h>
17 #include <linux/capability.h>
18 #include <linux/completion.h>
19 #include <linux/personality.h>
20 #include <linux/tty.h>
21 #include <linux/iocontext.h>
22 #include <linux/key.h>
23 #include <linux/cpu.h>
24 #include <linux/acct.h>
25 #include <linux/tsacct_kern.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/freezer.h>
29 #include <linux/binfmts.h>
30 #include <linux/nsproxy.h>
31 #include <linux/pid_namespace.h>
32 #include <linux/ptrace.h>
33 #include <linux/profile.h>
34 #include <linux/mount.h>
35 #include <linux/proc_fs.h>
36 #include <linux/kthread.h>
37 #include <linux/mempolicy.h>
38 #include <linux/taskstats_kern.h>
39 #include <linux/delayacct.h>
40 #include <linux/cgroup.h>
41 #include <linux/syscalls.h>
42 #include <linux/signal.h>
43 #include <linux/posix-timers.h>
44 #include <linux/cn_proc.h>
45 #include <linux/mutex.h>
46 #include <linux/futex.h>
47 #include <linux/pipe_fs_i.h>
48 #include <linux/audit.h> /* for audit_free() */
49 #include <linux/resource.h>
50 #include <linux/blkdev.h>
51 #include <linux/task_io_accounting_ops.h>
52 #include <linux/tracehook.h>
53 #include <linux/fs_struct.h>
54 #include <linux/init_task.h>
55 #include <linux/perf_event.h>
56 #include <trace/events/sched.h>
57 #include <linux/hw_breakpoint.h>
58 #include <linux/oom.h>
59 #include <linux/writeback.h>
60 #include <linux/shm.h>
61 #include <linux/kcov.h>
62 #include <linux/random.h>
63 #include <linux/rcuwait.h>
64 #include <linux/compat.h>
66 #include <linux/uaccess.h>
67 #include <asm/unistd.h>
68 #include <asm/pgtable.h>
69 #include <asm/mmu_context.h>
71 static void __unhash_process(struct task_struct
*p
, bool group_dead
)
74 detach_pid(p
, PIDTYPE_PID
);
76 detach_pid(p
, PIDTYPE_PGID
);
77 detach_pid(p
, PIDTYPE_SID
);
79 list_del_rcu(&p
->tasks
);
80 list_del_init(&p
->sibling
);
81 __this_cpu_dec(process_counts
);
83 list_del_rcu(&p
->thread_group
);
84 list_del_rcu(&p
->thread_node
);
88 * This function expects the tasklist_lock write-locked.
90 static void __exit_signal(struct task_struct
*tsk
)
92 struct signal_struct
*sig
= tsk
->signal
;
93 bool group_dead
= thread_group_leader(tsk
);
94 struct sighand_struct
*sighand
;
95 struct tty_struct
*uninitialized_var(tty
);
98 sighand
= rcu_dereference_check(tsk
->sighand
,
99 lockdep_tasklist_lock_is_held());
100 spin_lock(&sighand
->siglock
);
102 #ifdef CONFIG_POSIX_TIMERS
103 posix_cpu_timers_exit(tsk
);
105 posix_cpu_timers_exit_group(tsk
);
108 * This can only happen if the caller is de_thread().
109 * FIXME: this is the temporary hack, we should teach
110 * posix-cpu-timers to handle this case correctly.
112 if (unlikely(has_group_leader_pid(tsk
)))
113 posix_cpu_timers_exit_group(tsk
);
122 * If there is any task waiting for the group exit
125 if (sig
->notify_count
> 0 && !--sig
->notify_count
)
126 wake_up_process(sig
->group_exit_task
);
128 if (tsk
== sig
->curr_target
)
129 sig
->curr_target
= next_thread(tsk
);
132 add_device_randomness((const void*) &tsk
->se
.sum_exec_runtime
,
133 sizeof(unsigned long long));
136 * Accumulate here the counters for all threads as they die. We could
137 * skip the group leader because it is the last user of signal_struct,
138 * but we want to avoid the race with thread_group_cputime() which can
139 * see the empty ->thread_head list.
141 task_cputime(tsk
, &utime
, &stime
);
142 write_seqlock(&sig
->stats_lock
);
145 sig
->gtime
+= task_gtime(tsk
);
146 sig
->min_flt
+= tsk
->min_flt
;
147 sig
->maj_flt
+= tsk
->maj_flt
;
148 sig
->nvcsw
+= tsk
->nvcsw
;
149 sig
->nivcsw
+= tsk
->nivcsw
;
150 sig
->inblock
+= task_io_get_inblock(tsk
);
151 sig
->oublock
+= task_io_get_oublock(tsk
);
152 task_io_accounting_add(&sig
->ioac
, &tsk
->ioac
);
153 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
155 __unhash_process(tsk
, group_dead
);
156 write_sequnlock(&sig
->stats_lock
);
159 * Do this under ->siglock, we can race with another thread
160 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
162 flush_sigqueue(&tsk
->pending
);
164 spin_unlock(&sighand
->siglock
);
166 __cleanup_sighand(sighand
);
167 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
169 flush_sigqueue(&sig
->shared_pending
);
174 static void delayed_put_task_struct(struct rcu_head
*rhp
)
176 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
178 perf_event_delayed_put(tsk
);
179 trace_sched_process_free(tsk
);
180 put_task_struct(tsk
);
184 void release_task(struct task_struct
*p
)
186 struct task_struct
*leader
;
189 /* don't need to get the RCU readlock here - the process is dead and
190 * can't be modifying its own credentials. But shut RCU-lockdep up */
192 atomic_dec(&__task_cred(p
)->user
->processes
);
198 write_lock_irq(&tasklist_lock
);
199 ptrace_release_task(p
);
203 * If we are the last non-leader member of the thread
204 * group, and the leader is zombie, then notify the
205 * group leader's parent process. (if it wants notification.)
208 leader
= p
->group_leader
;
209 if (leader
!= p
&& thread_group_empty(leader
)
210 && leader
->exit_state
== EXIT_ZOMBIE
) {
212 * If we were the last child thread and the leader has
213 * exited already, and the leader's parent ignores SIGCHLD,
214 * then we are the one who should release the leader.
216 zap_leader
= do_notify_parent(leader
, leader
->exit_signal
);
218 leader
->exit_state
= EXIT_DEAD
;
221 write_unlock_irq(&tasklist_lock
);
223 call_rcu(&p
->rcu
, delayed_put_task_struct
);
226 if (unlikely(zap_leader
))
231 * Note that if this function returns a valid task_struct pointer (!NULL)
232 * task->usage must remain >0 for the duration of the RCU critical section.
234 struct task_struct
*task_rcu_dereference(struct task_struct
**ptask
)
236 struct sighand_struct
*sighand
;
237 struct task_struct
*task
;
240 * We need to verify that release_task() was not called and thus
241 * delayed_put_task_struct() can't run and drop the last reference
242 * before rcu_read_unlock(). We check task->sighand != NULL,
243 * but we can read the already freed and reused memory.
246 task
= rcu_dereference(*ptask
);
250 probe_kernel_address(&task
->sighand
, sighand
);
253 * Pairs with atomic_dec_and_test() in put_task_struct(). If this task
254 * was already freed we can not miss the preceding update of this
258 if (unlikely(task
!= READ_ONCE(*ptask
)))
262 * We've re-checked that "task == *ptask", now we have two different
265 * 1. This is actually the same task/task_struct. In this case
266 * sighand != NULL tells us it is still alive.
268 * 2. This is another task which got the same memory for task_struct.
269 * We can't know this of course, and we can not trust
272 * In this case we actually return a random value, but this is
275 * If we return NULL - we can pretend that we actually noticed that
276 * *ptask was updated when the previous task has exited. Or pretend
277 * that probe_slab_address(&sighand) reads NULL.
279 * If we return the new task (because sighand is not NULL for any
280 * reason) - this is fine too. This (new) task can't go away before
283 * And note: We could even eliminate the false positive if re-read
284 * task->sighand once again to avoid the falsely NULL. But this case
285 * is very unlikely so we don't care.
293 void rcuwait_wake_up(struct rcuwait
*w
)
295 struct task_struct
*task
;
300 * Order condition vs @task, such that everything prior to the load
301 * of @task is visible. This is the condition as to why the user called
302 * rcuwait_trywake() in the first place. Pairs with set_current_state()
303 * barrier (A) in rcuwait_wait_event().
306 * [S] tsk = current [S] cond = true
313 * Avoid using task_rcu_dereference() magic as long as we are careful,
314 * see comment in rcuwait_wait_event() regarding ->exit_state.
316 task
= rcu_dereference(w
->task
);
318 wake_up_process(task
);
323 * Determine if a process group is "orphaned", according to the POSIX
324 * definition in 2.2.2.52. Orphaned process groups are not to be affected
325 * by terminal-generated stop signals. Newly orphaned process groups are
326 * to receive a SIGHUP and a SIGCONT.
328 * "I ask you, have you ever known what it is to be an orphan?"
330 static int will_become_orphaned_pgrp(struct pid
*pgrp
,
331 struct task_struct
*ignored_task
)
333 struct task_struct
*p
;
335 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
336 if ((p
== ignored_task
) ||
337 (p
->exit_state
&& thread_group_empty(p
)) ||
338 is_global_init(p
->real_parent
))
341 if (task_pgrp(p
->real_parent
) != pgrp
&&
342 task_session(p
->real_parent
) == task_session(p
))
344 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
349 int is_current_pgrp_orphaned(void)
353 read_lock(&tasklist_lock
);
354 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
355 read_unlock(&tasklist_lock
);
360 static bool has_stopped_jobs(struct pid
*pgrp
)
362 struct task_struct
*p
;
364 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
365 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
367 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
373 * Check to see if any process groups have become orphaned as
374 * a result of our exiting, and if they have any stopped jobs,
375 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
378 kill_orphaned_pgrp(struct task_struct
*tsk
, struct task_struct
*parent
)
380 struct pid
*pgrp
= task_pgrp(tsk
);
381 struct task_struct
*ignored_task
= tsk
;
384 /* exit: our father is in a different pgrp than
385 * we are and we were the only connection outside.
387 parent
= tsk
->real_parent
;
389 /* reparent: our child is in a different pgrp than
390 * we are, and it was the only connection outside.
394 if (task_pgrp(parent
) != pgrp
&&
395 task_session(parent
) == task_session(tsk
) &&
396 will_become_orphaned_pgrp(pgrp
, ignored_task
) &&
397 has_stopped_jobs(pgrp
)) {
398 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
399 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
405 * A task is exiting. If it owned this mm, find a new owner for the mm.
407 void mm_update_next_owner(struct mm_struct
*mm
)
409 struct task_struct
*c
, *g
, *p
= current
;
413 * If the exiting or execing task is not the owner, it's
414 * someone else's problem.
419 * The current owner is exiting/execing and there are no other
420 * candidates. Do not leave the mm pointing to a possibly
421 * freed task structure.
423 if (atomic_read(&mm
->mm_users
) <= 1) {
428 read_lock(&tasklist_lock
);
430 * Search in the children
432 list_for_each_entry(c
, &p
->children
, sibling
) {
434 goto assign_new_owner
;
438 * Search in the siblings
440 list_for_each_entry(c
, &p
->real_parent
->children
, sibling
) {
442 goto assign_new_owner
;
446 * Search through everything else, we should not get here often.
448 for_each_process(g
) {
449 if (g
->flags
& PF_KTHREAD
)
451 for_each_thread(g
, c
) {
453 goto assign_new_owner
;
458 read_unlock(&tasklist_lock
);
460 * We found no owner yet mm_users > 1: this implies that we are
461 * most likely racing with swapoff (try_to_unuse()) or /proc or
462 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
471 * The task_lock protects c->mm from changing.
472 * We always want mm->owner->mm == mm
476 * Delay read_unlock() till we have the task_lock()
477 * to ensure that c does not slip away underneath us
479 read_unlock(&tasklist_lock
);
489 #endif /* CONFIG_MEMCG */
492 * Turn us into a lazy TLB process if we
495 static void exit_mm(void)
497 struct mm_struct
*mm
= current
->mm
;
498 struct core_state
*core_state
;
500 exit_mm_release(current
, mm
);
505 * Serialize with any possible pending coredump.
506 * We must hold mmap_sem around checking core_state
507 * and clearing tsk->mm. The core-inducing thread
508 * will increment ->nr_threads for each thread in the
509 * group with ->mm != NULL.
511 down_read(&mm
->mmap_sem
);
512 core_state
= mm
->core_state
;
514 struct core_thread self
;
516 up_read(&mm
->mmap_sem
);
519 self
.next
= xchg(&core_state
->dumper
.next
, &self
);
521 * Implies mb(), the result of xchg() must be visible
522 * to core_state->dumper.
524 if (atomic_dec_and_test(&core_state
->nr_threads
))
525 complete(&core_state
->startup
);
528 set_current_state(TASK_UNINTERRUPTIBLE
);
529 if (!self
.task
) /* see coredump_finish() */
531 freezable_schedule();
533 __set_current_state(TASK_RUNNING
);
534 down_read(&mm
->mmap_sem
);
537 BUG_ON(mm
!= current
->active_mm
);
538 /* more a memory barrier than a real lock */
541 up_read(&mm
->mmap_sem
);
542 enter_lazy_tlb(mm
, current
);
543 task_unlock(current
);
544 mm_update_next_owner(mm
);
546 if (test_thread_flag(TIF_MEMDIE
))
550 static struct task_struct
*find_alive_thread(struct task_struct
*p
)
552 struct task_struct
*t
;
554 for_each_thread(p
, t
) {
555 if (!(t
->flags
& PF_EXITING
))
561 static struct task_struct
*find_child_reaper(struct task_struct
*father
,
562 struct list_head
*dead
)
563 __releases(&tasklist_lock
)
564 __acquires(&tasklist_lock
)
566 struct pid_namespace
*pid_ns
= task_active_pid_ns(father
);
567 struct task_struct
*reaper
= pid_ns
->child_reaper
;
568 struct task_struct
*p
, *n
;
570 if (likely(reaper
!= father
))
573 reaper
= find_alive_thread(father
);
575 pid_ns
->child_reaper
= reaper
;
579 write_unlock_irq(&tasklist_lock
);
581 list_for_each_entry_safe(p
, n
, dead
, ptrace_entry
) {
582 list_del_init(&p
->ptrace_entry
);
586 zap_pid_ns_processes(pid_ns
);
587 write_lock_irq(&tasklist_lock
);
593 * When we die, we re-parent all our children, and try to:
594 * 1. give them to another thread in our thread group, if such a member exists
595 * 2. give it to the first ancestor process which prctl'd itself as a
596 * child_subreaper for its children (like a service manager)
597 * 3. give it to the init process (PID 1) in our pid namespace
599 static struct task_struct
*find_new_reaper(struct task_struct
*father
,
600 struct task_struct
*child_reaper
)
602 struct task_struct
*thread
, *reaper
;
604 thread
= find_alive_thread(father
);
608 if (father
->signal
->has_child_subreaper
) {
609 unsigned int ns_level
= task_pid(father
)->level
;
611 * Find the first ->is_child_subreaper ancestor in our pid_ns.
612 * We can't check reaper != child_reaper to ensure we do not
613 * cross the namespaces, the exiting parent could be injected
614 * by setns() + fork().
615 * We check pid->level, this is slightly more efficient than
616 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
618 for (reaper
= father
->real_parent
;
619 task_pid(reaper
)->level
== ns_level
;
620 reaper
= reaper
->real_parent
) {
621 if (reaper
== &init_task
)
623 if (!reaper
->signal
->is_child_subreaper
)
625 thread
= find_alive_thread(reaper
);
635 * Any that need to be release_task'd are put on the @dead list.
637 static void reparent_leader(struct task_struct
*father
, struct task_struct
*p
,
638 struct list_head
*dead
)
640 if (unlikely(p
->exit_state
== EXIT_DEAD
))
643 /* We don't want people slaying init. */
644 p
->exit_signal
= SIGCHLD
;
646 /* If it has exited notify the new parent about this child's death. */
648 p
->exit_state
== EXIT_ZOMBIE
&& thread_group_empty(p
)) {
649 if (do_notify_parent(p
, p
->exit_signal
)) {
650 p
->exit_state
= EXIT_DEAD
;
651 list_add(&p
->ptrace_entry
, dead
);
655 kill_orphaned_pgrp(p
, father
);
659 * This does two things:
661 * A. Make init inherit all the child processes
662 * B. Check to see if any process groups have become orphaned
663 * as a result of our exiting, and if they have any stopped
664 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
666 static void forget_original_parent(struct task_struct
*father
,
667 struct list_head
*dead
)
669 struct task_struct
*p
, *t
, *reaper
;
671 if (unlikely(!list_empty(&father
->ptraced
)))
672 exit_ptrace(father
, dead
);
674 /* Can drop and reacquire tasklist_lock */
675 reaper
= find_child_reaper(father
, dead
);
676 if (list_empty(&father
->children
))
679 reaper
= find_new_reaper(father
, reaper
);
680 list_for_each_entry(p
, &father
->children
, sibling
) {
681 for_each_thread(p
, t
) {
682 t
->real_parent
= reaper
;
683 BUG_ON((!t
->ptrace
) != (t
->parent
== father
));
684 if (likely(!t
->ptrace
))
685 t
->parent
= t
->real_parent
;
686 if (t
->pdeath_signal
)
687 group_send_sig_info(t
->pdeath_signal
,
691 * If this is a threaded reparent there is no need to
692 * notify anyone anything has happened.
694 if (!same_thread_group(reaper
, father
))
695 reparent_leader(father
, p
, dead
);
697 list_splice_tail_init(&father
->children
, &reaper
->children
);
701 * Send signals to all our closest relatives so that they know
702 * to properly mourn us..
704 static void exit_notify(struct task_struct
*tsk
, int group_dead
)
707 struct task_struct
*p
, *n
;
710 write_lock_irq(&tasklist_lock
);
711 forget_original_parent(tsk
, &dead
);
714 kill_orphaned_pgrp(tsk
->group_leader
, NULL
);
716 if (unlikely(tsk
->ptrace
)) {
717 int sig
= thread_group_leader(tsk
) &&
718 thread_group_empty(tsk
) &&
719 !ptrace_reparented(tsk
) ?
720 tsk
->exit_signal
: SIGCHLD
;
721 autoreap
= do_notify_parent(tsk
, sig
);
722 } else if (thread_group_leader(tsk
)) {
723 autoreap
= thread_group_empty(tsk
) &&
724 do_notify_parent(tsk
, tsk
->exit_signal
);
729 tsk
->exit_state
= autoreap
? EXIT_DEAD
: EXIT_ZOMBIE
;
730 if (tsk
->exit_state
== EXIT_DEAD
)
731 list_add(&tsk
->ptrace_entry
, &dead
);
733 /* mt-exec, de_thread() is waiting for group leader */
734 if (unlikely(tsk
->signal
->notify_count
< 0))
735 wake_up_process(tsk
->signal
->group_exit_task
);
736 write_unlock_irq(&tasklist_lock
);
738 list_for_each_entry_safe(p
, n
, &dead
, ptrace_entry
) {
739 list_del_init(&p
->ptrace_entry
);
744 #ifdef CONFIG_DEBUG_STACK_USAGE
745 static void check_stack_usage(void)
747 static DEFINE_SPINLOCK(low_water_lock
);
748 static int lowest_to_date
= THREAD_SIZE
;
751 free
= stack_not_used(current
);
753 if (free
>= lowest_to_date
)
756 spin_lock(&low_water_lock
);
757 if (free
< lowest_to_date
) {
758 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
759 current
->comm
, task_pid_nr(current
), free
);
760 lowest_to_date
= free
;
762 spin_unlock(&low_water_lock
);
765 static inline void check_stack_usage(void) {}
768 void __noreturn
do_exit(long code
)
770 struct task_struct
*tsk
= current
;
773 profile_task_exit(tsk
);
776 WARN_ON(blk_needs_flush_plug(tsk
));
778 if (unlikely(in_interrupt()))
779 panic("Aiee, killing interrupt handler!");
780 if (unlikely(!tsk
->pid
))
781 panic("Attempted to kill the idle task!");
784 * If do_exit is called because this processes oopsed, it's possible
785 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
786 * continuing. Amongst other possible reasons, this is to prevent
787 * mm_release()->clear_child_tid() from writing to a user-controlled
792 ptrace_event(PTRACE_EVENT_EXIT
, code
);
794 validate_creds_for_do_exit(tsk
);
797 * We're taking recursive faults here in do_exit. Safest is to just
798 * leave this task alone and wait for reboot.
800 if (unlikely(tsk
->flags
& PF_EXITING
)) {
801 pr_alert("Fixing recursive fault but reboot is needed!\n");
802 futex_exit_recursive(tsk
);
803 set_current_state(TASK_UNINTERRUPTIBLE
);
807 exit_signals(tsk
); /* sets PF_EXITING */
809 if (unlikely(in_atomic())) {
810 pr_info("note: %s[%d] exited with preempt_count %d\n",
811 current
->comm
, task_pid_nr(current
),
813 preempt_count_set(PREEMPT_ENABLED
);
816 /* sync mm's RSS info before statistics gathering */
818 sync_mm_rss(tsk
->mm
);
819 acct_update_integrals(tsk
);
820 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
823 * If the last thread of global init has exited, panic
824 * immediately to get a useable coredump.
826 if (unlikely(is_global_init(tsk
)))
827 panic("Attempted to kill init! exitcode=0x%08x\n",
828 tsk
->signal
->group_exit_code
?: (int)code
);
830 #ifdef CONFIG_POSIX_TIMERS
831 hrtimer_cancel(&tsk
->signal
->real_timer
);
832 exit_itimers(tsk
->signal
);
835 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, tsk
->mm
);
837 acct_collect(code
, group_dead
);
842 tsk
->exit_code
= code
;
843 taskstats_exit(tsk
, group_dead
);
849 trace_sched_process_exit(tsk
);
856 disassociate_ctty(1);
857 exit_task_namespaces(tsk
);
862 * Flush inherited counters to the parent - before the parent
863 * gets woken up by child-exit notifications.
865 * because of cgroup mode, must be called before cgroup_exit()
867 perf_event_exit_task(tsk
);
869 sched_autogroup_exit_task(tsk
);
873 * FIXME: do that only when needed, using sched_exit tracepoint
875 flush_ptrace_hw_breakpoint(tsk
);
877 exit_tasks_rcu_start();
878 exit_notify(tsk
, group_dead
);
879 proc_exit_connector(tsk
);
880 mpol_put_task_policy(tsk
);
882 if (unlikely(current
->pi_state_cache
))
883 kfree(current
->pi_state_cache
);
886 * Make sure we are holding no locks:
888 debug_check_no_locks_held();
891 exit_io_context(tsk
);
893 if (tsk
->splice_pipe
)
894 free_pipe_info(tsk
->splice_pipe
);
896 if (tsk
->task_frag
.page
)
897 put_page(tsk
->task_frag
.page
);
899 validate_creds_for_do_exit(tsk
);
904 __this_cpu_add(dirty_throttle_leaks
, tsk
->nr_dirtied
);
906 exit_tasks_rcu_finish();
908 lockdep_free_task(tsk
);
911 EXPORT_SYMBOL_GPL(do_exit
);
913 void complete_and_exit(struct completion
*comp
, long code
)
920 EXPORT_SYMBOL(complete_and_exit
);
922 SYSCALL_DEFINE1(exit
, int, error_code
)
924 do_exit((error_code
&0xff)<<8);
928 * Take down every thread in the group. This is called by fatal signals
929 * as well as by sys_exit_group (below).
932 do_group_exit(int exit_code
)
934 struct signal_struct
*sig
= current
->signal
;
936 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
938 if (signal_group_exit(sig
))
939 exit_code
= sig
->group_exit_code
;
940 else if (!thread_group_empty(current
)) {
941 struct sighand_struct
*const sighand
= current
->sighand
;
943 spin_lock_irq(&sighand
->siglock
);
944 if (signal_group_exit(sig
))
945 /* Another thread got here before we took the lock. */
946 exit_code
= sig
->group_exit_code
;
948 sig
->group_exit_code
= exit_code
;
949 sig
->flags
= SIGNAL_GROUP_EXIT
;
950 zap_other_threads(current
);
952 spin_unlock_irq(&sighand
->siglock
);
960 * this kills every thread in the thread group. Note that any externally
961 * wait4()-ing process will get the correct exit code - even if this
962 * thread is not the thread group leader.
964 SYSCALL_DEFINE1(exit_group
, int, error_code
)
966 do_group_exit((error_code
& 0xff) << 8);
979 enum pid_type wo_type
;
983 struct waitid_info
*wo_info
;
985 struct rusage
*wo_rusage
;
987 wait_queue_entry_t child_wait
;
992 struct pid
*task_pid_type(struct task_struct
*task
, enum pid_type type
)
994 if (type
!= PIDTYPE_PID
)
995 task
= task
->group_leader
;
996 return task
->pids
[type
].pid
;
999 static int eligible_pid(struct wait_opts
*wo
, struct task_struct
*p
)
1001 return wo
->wo_type
== PIDTYPE_MAX
||
1002 task_pid_type(p
, wo
->wo_type
) == wo
->wo_pid
;
1006 eligible_child(struct wait_opts
*wo
, bool ptrace
, struct task_struct
*p
)
1008 if (!eligible_pid(wo
, p
))
1012 * Wait for all children (clone and not) if __WALL is set or
1013 * if it is traced by us.
1015 if (ptrace
|| (wo
->wo_flags
& __WALL
))
1019 * Otherwise, wait for clone children *only* if __WCLONE is set;
1020 * otherwise, wait for non-clone children *only*.
1022 * Note: a "clone" child here is one that reports to its parent
1023 * using a signal other than SIGCHLD, or a non-leader thread which
1024 * we can only see if it is traced by us.
1026 if ((p
->exit_signal
!= SIGCHLD
) ^ !!(wo
->wo_flags
& __WCLONE
))
1033 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1034 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1035 * the lock and this task is uninteresting. If we return nonzero, we have
1036 * released the lock and the system call should return.
1038 static int wait_task_zombie(struct wait_opts
*wo
, struct task_struct
*p
)
1041 pid_t pid
= task_pid_vnr(p
);
1042 uid_t uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1043 struct waitid_info
*infop
;
1045 if (!likely(wo
->wo_flags
& WEXITED
))
1048 if (unlikely(wo
->wo_flags
& WNOWAIT
)) {
1049 status
= p
->exit_code
;
1051 read_unlock(&tasklist_lock
);
1052 sched_annotate_sleep();
1054 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1059 * Move the task's state to DEAD/TRACE, only one thread can do this.
1061 state
= (ptrace_reparented(p
) && thread_group_leader(p
)) ?
1062 EXIT_TRACE
: EXIT_DEAD
;
1063 if (cmpxchg(&p
->exit_state
, EXIT_ZOMBIE
, state
) != EXIT_ZOMBIE
)
1066 * We own this thread, nobody else can reap it.
1068 read_unlock(&tasklist_lock
);
1069 sched_annotate_sleep();
1072 * Check thread_group_leader() to exclude the traced sub-threads.
1074 if (state
== EXIT_DEAD
&& thread_group_leader(p
)) {
1075 struct signal_struct
*sig
= p
->signal
;
1076 struct signal_struct
*psig
= current
->signal
;
1077 unsigned long maxrss
;
1078 u64 tgutime
, tgstime
;
1081 * The resource counters for the group leader are in its
1082 * own task_struct. Those for dead threads in the group
1083 * are in its signal_struct, as are those for the child
1084 * processes it has previously reaped. All these
1085 * accumulate in the parent's signal_struct c* fields.
1087 * We don't bother to take a lock here to protect these
1088 * p->signal fields because the whole thread group is dead
1089 * and nobody can change them.
1091 * psig->stats_lock also protects us from our sub-theads
1092 * which can reap other children at the same time. Until
1093 * we change k_getrusage()-like users to rely on this lock
1094 * we have to take ->siglock as well.
1096 * We use thread_group_cputime_adjusted() to get times for
1097 * the thread group, which consolidates times for all threads
1098 * in the group including the group leader.
1100 thread_group_cputime_adjusted(p
, &tgutime
, &tgstime
);
1101 spin_lock_irq(¤t
->sighand
->siglock
);
1102 write_seqlock(&psig
->stats_lock
);
1103 psig
->cutime
+= tgutime
+ sig
->cutime
;
1104 psig
->cstime
+= tgstime
+ sig
->cstime
;
1105 psig
->cgtime
+= task_gtime(p
) + sig
->gtime
+ sig
->cgtime
;
1107 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1109 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1111 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1113 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1115 task_io_get_inblock(p
) +
1116 sig
->inblock
+ sig
->cinblock
;
1118 task_io_get_oublock(p
) +
1119 sig
->oublock
+ sig
->coublock
;
1120 maxrss
= max(sig
->maxrss
, sig
->cmaxrss
);
1121 if (psig
->cmaxrss
< maxrss
)
1122 psig
->cmaxrss
= maxrss
;
1123 task_io_accounting_add(&psig
->ioac
, &p
->ioac
);
1124 task_io_accounting_add(&psig
->ioac
, &sig
->ioac
);
1125 write_sequnlock(&psig
->stats_lock
);
1126 spin_unlock_irq(¤t
->sighand
->siglock
);
1130 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1131 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1132 ? p
->signal
->group_exit_code
: p
->exit_code
;
1133 wo
->wo_stat
= status
;
1135 if (state
== EXIT_TRACE
) {
1136 write_lock_irq(&tasklist_lock
);
1137 /* We dropped tasklist, ptracer could die and untrace */
1140 /* If parent wants a zombie, don't release it now */
1141 state
= EXIT_ZOMBIE
;
1142 if (do_notify_parent(p
, p
->exit_signal
))
1144 p
->exit_state
= state
;
1145 write_unlock_irq(&tasklist_lock
);
1147 if (state
== EXIT_DEAD
)
1151 infop
= wo
->wo_info
;
1153 if ((status
& 0x7f) == 0) {
1154 infop
->cause
= CLD_EXITED
;
1155 infop
->status
= status
>> 8;
1157 infop
->cause
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1158 infop
->status
= status
& 0x7f;
1167 static int *task_stopped_code(struct task_struct
*p
, bool ptrace
)
1170 if (task_is_traced(p
) && !(p
->jobctl
& JOBCTL_LISTENING
))
1171 return &p
->exit_code
;
1173 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
1174 return &p
->signal
->group_exit_code
;
1180 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1182 * @ptrace: is the wait for ptrace
1183 * @p: task to wait for
1185 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1188 * read_lock(&tasklist_lock), which is released if return value is
1189 * non-zero. Also, grabs and releases @p->sighand->siglock.
1192 * 0 if wait condition didn't exist and search for other wait conditions
1193 * should continue. Non-zero return, -errno on failure and @p's pid on
1194 * success, implies that tasklist_lock is released and wait condition
1195 * search should terminate.
1197 static int wait_task_stopped(struct wait_opts
*wo
,
1198 int ptrace
, struct task_struct
*p
)
1200 struct waitid_info
*infop
;
1201 int exit_code
, *p_code
, why
;
1202 uid_t uid
= 0; /* unneeded, required by compiler */
1206 * Traditionally we see ptrace'd stopped tasks regardless of options.
1208 if (!ptrace
&& !(wo
->wo_flags
& WUNTRACED
))
1211 if (!task_stopped_code(p
, ptrace
))
1215 spin_lock_irq(&p
->sighand
->siglock
);
1217 p_code
= task_stopped_code(p
, ptrace
);
1218 if (unlikely(!p_code
))
1221 exit_code
= *p_code
;
1225 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1228 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1230 spin_unlock_irq(&p
->sighand
->siglock
);
1235 * Now we are pretty sure this task is interesting.
1236 * Make sure it doesn't get reaped out from under us while we
1237 * give up the lock and then examine it below. We don't want to
1238 * keep holding onto the tasklist_lock while we call getrusage and
1239 * possibly take page faults for user memory.
1242 pid
= task_pid_vnr(p
);
1243 why
= ptrace
? CLD_TRAPPED
: CLD_STOPPED
;
1244 read_unlock(&tasklist_lock
);
1245 sched_annotate_sleep();
1247 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1250 if (likely(!(wo
->wo_flags
& WNOWAIT
)))
1251 wo
->wo_stat
= (exit_code
<< 8) | 0x7f;
1253 infop
= wo
->wo_info
;
1256 infop
->status
= exit_code
;
1264 * Handle do_wait work for one task in a live, non-stopped state.
1265 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1266 * the lock and this task is uninteresting. If we return nonzero, we have
1267 * released the lock and the system call should return.
1269 static int wait_task_continued(struct wait_opts
*wo
, struct task_struct
*p
)
1271 struct waitid_info
*infop
;
1275 if (!unlikely(wo
->wo_flags
& WCONTINUED
))
1278 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1281 spin_lock_irq(&p
->sighand
->siglock
);
1282 /* Re-check with the lock held. */
1283 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1284 spin_unlock_irq(&p
->sighand
->siglock
);
1287 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1288 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1289 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1290 spin_unlock_irq(&p
->sighand
->siglock
);
1292 pid
= task_pid_vnr(p
);
1294 read_unlock(&tasklist_lock
);
1295 sched_annotate_sleep();
1297 getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
);
1300 infop
= wo
->wo_info
;
1302 wo
->wo_stat
= 0xffff;
1304 infop
->cause
= CLD_CONTINUED
;
1307 infop
->status
= SIGCONT
;
1313 * Consider @p for a wait by @parent.
1315 * -ECHILD should be in ->notask_error before the first call.
1316 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1317 * Returns zero if the search for a child should continue;
1318 * then ->notask_error is 0 if @p is an eligible child,
1321 static int wait_consider_task(struct wait_opts
*wo
, int ptrace
,
1322 struct task_struct
*p
)
1325 * We can race with wait_task_zombie() from another thread.
1326 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1327 * can't confuse the checks below.
1329 int exit_state
= READ_ONCE(p
->exit_state
);
1332 if (unlikely(exit_state
== EXIT_DEAD
))
1335 ret
= eligible_child(wo
, ptrace
, p
);
1339 if (unlikely(exit_state
== EXIT_TRACE
)) {
1341 * ptrace == 0 means we are the natural parent. In this case
1342 * we should clear notask_error, debugger will notify us.
1344 if (likely(!ptrace
))
1345 wo
->notask_error
= 0;
1349 if (likely(!ptrace
) && unlikely(p
->ptrace
)) {
1351 * If it is traced by its real parent's group, just pretend
1352 * the caller is ptrace_do_wait() and reap this child if it
1355 * This also hides group stop state from real parent; otherwise
1356 * a single stop can be reported twice as group and ptrace stop.
1357 * If a ptracer wants to distinguish these two events for its
1358 * own children it should create a separate process which takes
1359 * the role of real parent.
1361 if (!ptrace_reparented(p
))
1366 if (exit_state
== EXIT_ZOMBIE
) {
1367 /* we don't reap group leaders with subthreads */
1368 if (!delay_group_leader(p
)) {
1370 * A zombie ptracee is only visible to its ptracer.
1371 * Notification and reaping will be cascaded to the
1372 * real parent when the ptracer detaches.
1374 if (unlikely(ptrace
) || likely(!p
->ptrace
))
1375 return wait_task_zombie(wo
, p
);
1379 * Allow access to stopped/continued state via zombie by
1380 * falling through. Clearing of notask_error is complex.
1384 * If WEXITED is set, notask_error should naturally be
1385 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1386 * so, if there are live subthreads, there are events to
1387 * wait for. If all subthreads are dead, it's still safe
1388 * to clear - this function will be called again in finite
1389 * amount time once all the subthreads are released and
1390 * will then return without clearing.
1394 * Stopped state is per-task and thus can't change once the
1395 * target task dies. Only continued and exited can happen.
1396 * Clear notask_error if WCONTINUED | WEXITED.
1398 if (likely(!ptrace
) || (wo
->wo_flags
& (WCONTINUED
| WEXITED
)))
1399 wo
->notask_error
= 0;
1402 * @p is alive and it's gonna stop, continue or exit, so
1403 * there always is something to wait for.
1405 wo
->notask_error
= 0;
1409 * Wait for stopped. Depending on @ptrace, different stopped state
1410 * is used and the two don't interact with each other.
1412 ret
= wait_task_stopped(wo
, ptrace
, p
);
1417 * Wait for continued. There's only one continued state and the
1418 * ptracer can consume it which can confuse the real parent. Don't
1419 * use WCONTINUED from ptracer. You don't need or want it.
1421 return wait_task_continued(wo
, p
);
1425 * Do the work of do_wait() for one thread in the group, @tsk.
1427 * -ECHILD should be in ->notask_error before the first call.
1428 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1429 * Returns zero if the search for a child should continue; then
1430 * ->notask_error is 0 if there were any eligible children,
1433 static int do_wait_thread(struct wait_opts
*wo
, struct task_struct
*tsk
)
1435 struct task_struct
*p
;
1437 list_for_each_entry(p
, &tsk
->children
, sibling
) {
1438 int ret
= wait_consider_task(wo
, 0, p
);
1447 static int ptrace_do_wait(struct wait_opts
*wo
, struct task_struct
*tsk
)
1449 struct task_struct
*p
;
1451 list_for_each_entry(p
, &tsk
->ptraced
, ptrace_entry
) {
1452 int ret
= wait_consider_task(wo
, 1, p
);
1461 static int child_wait_callback(wait_queue_entry_t
*wait
, unsigned mode
,
1462 int sync
, void *key
)
1464 struct wait_opts
*wo
= container_of(wait
, struct wait_opts
,
1466 struct task_struct
*p
= key
;
1468 if (!eligible_pid(wo
, p
))
1471 if ((wo
->wo_flags
& __WNOTHREAD
) && wait
->private != p
->parent
)
1474 return default_wake_function(wait
, mode
, sync
, key
);
1477 void __wake_up_parent(struct task_struct
*p
, struct task_struct
*parent
)
1479 __wake_up_sync_key(&parent
->signal
->wait_chldexit
,
1480 TASK_INTERRUPTIBLE
, 1, p
);
1483 static long do_wait(struct wait_opts
*wo
)
1485 struct task_struct
*tsk
;
1488 trace_sched_process_wait(wo
->wo_pid
);
1490 init_waitqueue_func_entry(&wo
->child_wait
, child_wait_callback
);
1491 wo
->child_wait
.private = current
;
1492 add_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1495 * If there is nothing that can match our criteria, just get out.
1496 * We will clear ->notask_error to zero if we see any child that
1497 * might later match our criteria, even if we are not able to reap
1500 wo
->notask_error
= -ECHILD
;
1501 if ((wo
->wo_type
< PIDTYPE_MAX
) &&
1502 (!wo
->wo_pid
|| hlist_empty(&wo
->wo_pid
->tasks
[wo
->wo_type
])))
1505 set_current_state(TASK_INTERRUPTIBLE
);
1506 read_lock(&tasklist_lock
);
1509 retval
= do_wait_thread(wo
, tsk
);
1513 retval
= ptrace_do_wait(wo
, tsk
);
1517 if (wo
->wo_flags
& __WNOTHREAD
)
1519 } while_each_thread(current
, tsk
);
1520 read_unlock(&tasklist_lock
);
1523 retval
= wo
->notask_error
;
1524 if (!retval
&& !(wo
->wo_flags
& WNOHANG
)) {
1525 retval
= -ERESTARTSYS
;
1526 if (!signal_pending(current
)) {
1532 __set_current_state(TASK_RUNNING
);
1533 remove_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1537 static long kernel_waitid(int which
, pid_t upid
, struct waitid_info
*infop
,
1538 int options
, struct rusage
*ru
)
1540 struct wait_opts wo
;
1541 struct pid
*pid
= NULL
;
1545 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
|
1546 __WNOTHREAD
|__WCLONE
|__WALL
))
1548 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1561 type
= PIDTYPE_PGID
;
1569 if (type
< PIDTYPE_MAX
)
1570 pid
= find_get_pid(upid
);
1574 wo
.wo_flags
= options
;
1583 SYSCALL_DEFINE5(waitid
, int, which
, pid_t
, upid
, struct siginfo __user
*,
1584 infop
, int, options
, struct rusage __user
*, ru
)
1587 struct waitid_info info
= {.status
= 0};
1588 long err
= kernel_waitid(which
, upid
, &info
, options
, ru
? &r
: NULL
);
1594 if (ru
&& copy_to_user(ru
, &r
, sizeof(struct rusage
)))
1600 if (!access_ok(VERIFY_WRITE
, infop
, sizeof(*infop
)))
1603 user_access_begin();
1604 unsafe_put_user(signo
, &infop
->si_signo
, Efault
);
1605 unsafe_put_user(0, &infop
->si_errno
, Efault
);
1606 unsafe_put_user(info
.cause
, &infop
->si_code
, Efault
);
1607 unsafe_put_user(info
.pid
, &infop
->si_pid
, Efault
);
1608 unsafe_put_user(info
.uid
, &infop
->si_uid
, Efault
);
1609 unsafe_put_user(info
.status
, &infop
->si_status
, Efault
);
1617 long kernel_wait4(pid_t upid
, int __user
*stat_addr
, int options
,
1620 struct wait_opts wo
;
1621 struct pid
*pid
= NULL
;
1625 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1626 __WNOTHREAD
|__WCLONE
|__WALL
))
1629 /* -INT_MIN is not defined */
1630 if (upid
== INT_MIN
)
1635 else if (upid
< 0) {
1636 type
= PIDTYPE_PGID
;
1637 pid
= find_get_pid(-upid
);
1638 } else if (upid
== 0) {
1639 type
= PIDTYPE_PGID
;
1640 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1641 } else /* upid > 0 */ {
1643 pid
= find_get_pid(upid
);
1648 wo
.wo_flags
= options
| WEXITED
;
1654 if (ret
> 0 && stat_addr
&& put_user(wo
.wo_stat
, stat_addr
))
1660 SYSCALL_DEFINE4(wait4
, pid_t
, upid
, int __user
*, stat_addr
,
1661 int, options
, struct rusage __user
*, ru
)
1664 long err
= kernel_wait4(upid
, stat_addr
, options
, ru
? &r
: NULL
);
1667 if (ru
&& copy_to_user(ru
, &r
, sizeof(struct rusage
)))
1673 #ifdef __ARCH_WANT_SYS_WAITPID
1676 * sys_waitpid() remains for compatibility. waitpid() should be
1677 * implemented by calling sys_wait4() from libc.a.
1679 SYSCALL_DEFINE3(waitpid
, pid_t
, pid
, int __user
*, stat_addr
, int, options
)
1681 return sys_wait4(pid
, stat_addr
, options
, NULL
);
1686 #ifdef CONFIG_COMPAT
1687 COMPAT_SYSCALL_DEFINE4(wait4
,
1689 compat_uint_t __user
*, stat_addr
,
1691 struct compat_rusage __user
*, ru
)
1694 long err
= kernel_wait4(pid
, stat_addr
, options
, ru
? &r
: NULL
);
1696 if (ru
&& put_compat_rusage(&r
, ru
))
1702 COMPAT_SYSCALL_DEFINE5(waitid
,
1703 int, which
, compat_pid_t
, pid
,
1704 struct compat_siginfo __user
*, infop
, int, options
,
1705 struct compat_rusage __user
*, uru
)
1708 struct waitid_info info
= {.status
= 0};
1709 long err
= kernel_waitid(which
, pid
, &info
, options
, uru
? &ru
: NULL
);
1715 /* kernel_waitid() overwrites everything in ru */
1716 if (COMPAT_USE_64BIT_TIME
)
1717 err
= copy_to_user(uru
, &ru
, sizeof(ru
));
1719 err
= put_compat_rusage(&ru
, uru
);
1728 if (!access_ok(VERIFY_WRITE
, infop
, sizeof(*infop
)))
1731 user_access_begin();
1732 unsafe_put_user(signo
, &infop
->si_signo
, Efault
);
1733 unsafe_put_user(0, &infop
->si_errno
, Efault
);
1734 unsafe_put_user(info
.cause
, &infop
->si_code
, Efault
);
1735 unsafe_put_user(info
.pid
, &infop
->si_pid
, Efault
);
1736 unsafe_put_user(info
.uid
, &infop
->si_uid
, Efault
);
1737 unsafe_put_user(info
.status
, &infop
->si_status
, Efault
);
1746 __weak
void abort(void)
1750 /* if that doesn't kill us, halt */
1751 panic("Oops failed to kill thread");
1753 EXPORT_SYMBOL(abort
);