]>
git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - kernel/exit.c
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>
57 #include <asm/uaccess.h>
58 #include <asm/unistd.h>
59 #include <asm/pgtable.h>
60 #include <asm/mmu_context.h>
62 static void exit_mm(struct task_struct
*tsk
);
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 posix_cpu_timers_exit(tsk
);
97 posix_cpu_timers_exit_group(tsk
);
102 * This can only happen if the caller is de_thread().
103 * FIXME: this is the temporary hack, we should teach
104 * posix-cpu-timers to handle this case correctly.
106 if (unlikely(has_group_leader_pid(tsk
)))
107 posix_cpu_timers_exit_group(tsk
);
110 * If there is any task waiting for the group exit
113 if (sig
->notify_count
> 0 && !--sig
->notify_count
)
114 wake_up_process(sig
->group_exit_task
);
116 if (tsk
== sig
->curr_target
)
117 sig
->curr_target
= next_thread(tsk
);
121 * Accumulate here the counters for all threads as they die. We could
122 * skip the group leader because it is the last user of signal_struct,
123 * but we want to avoid the race with thread_group_cputime() which can
124 * see the empty ->thread_head list.
126 task_cputime(tsk
, &utime
, &stime
);
127 write_seqlock(&sig
->stats_lock
);
130 sig
->gtime
+= task_gtime(tsk
);
131 sig
->min_flt
+= tsk
->min_flt
;
132 sig
->maj_flt
+= tsk
->maj_flt
;
133 sig
->nvcsw
+= tsk
->nvcsw
;
134 sig
->nivcsw
+= tsk
->nivcsw
;
135 sig
->inblock
+= task_io_get_inblock(tsk
);
136 sig
->oublock
+= task_io_get_oublock(tsk
);
137 task_io_accounting_add(&sig
->ioac
, &tsk
->ioac
);
138 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
140 __unhash_process(tsk
, group_dead
);
141 write_sequnlock(&sig
->stats_lock
);
144 * Do this under ->siglock, we can race with another thread
145 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
147 flush_sigqueue(&tsk
->pending
);
149 spin_unlock(&sighand
->siglock
);
151 __cleanup_sighand(sighand
);
152 clear_tsk_thread_flag(tsk
, TIF_SIGPENDING
);
154 flush_sigqueue(&sig
->shared_pending
);
159 static void delayed_put_task_struct(struct rcu_head
*rhp
)
161 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
163 perf_event_delayed_put(tsk
);
164 trace_sched_process_free(tsk
);
165 put_task_struct(tsk
);
169 void release_task(struct task_struct
*p
)
171 struct task_struct
*leader
;
174 /* don't need to get the RCU readlock here - the process is dead and
175 * can't be modifying its own credentials. But shut RCU-lockdep up */
177 atomic_dec(&__task_cred(p
)->user
->processes
);
182 write_lock_irq(&tasklist_lock
);
183 ptrace_release_task(p
);
187 * If we are the last non-leader member of the thread
188 * group, and the leader is zombie, then notify the
189 * group leader's parent process. (if it wants notification.)
192 leader
= p
->group_leader
;
193 if (leader
!= p
&& thread_group_empty(leader
)
194 && leader
->exit_state
== EXIT_ZOMBIE
) {
196 * If we were the last child thread and the leader has
197 * exited already, and the leader's parent ignores SIGCHLD,
198 * then we are the one who should release the leader.
200 zap_leader
= do_notify_parent(leader
, leader
->exit_signal
);
202 leader
->exit_state
= EXIT_DEAD
;
205 write_unlock_irq(&tasklist_lock
);
207 call_rcu(&p
->rcu
, delayed_put_task_struct
);
210 if (unlikely(zap_leader
))
215 * This checks not only the pgrp, but falls back on the pid if no
216 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
219 * The caller must hold rcu lock or the tasklist lock.
221 struct pid
*session_of_pgrp(struct pid
*pgrp
)
223 struct task_struct
*p
;
224 struct pid
*sid
= NULL
;
226 p
= pid_task(pgrp
, PIDTYPE_PGID
);
228 p
= pid_task(pgrp
, PIDTYPE_PID
);
230 sid
= task_session(p
);
236 * Determine if a process group is "orphaned", according to the POSIX
237 * definition in 2.2.2.52. Orphaned process groups are not to be affected
238 * by terminal-generated stop signals. Newly orphaned process groups are
239 * to receive a SIGHUP and a SIGCONT.
241 * "I ask you, have you ever known what it is to be an orphan?"
243 static int will_become_orphaned_pgrp(struct pid
*pgrp
,
244 struct task_struct
*ignored_task
)
246 struct task_struct
*p
;
248 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
249 if ((p
== ignored_task
) ||
250 (p
->exit_state
&& thread_group_empty(p
)) ||
251 is_global_init(p
->real_parent
))
254 if (task_pgrp(p
->real_parent
) != pgrp
&&
255 task_session(p
->real_parent
) == task_session(p
))
257 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
262 int is_current_pgrp_orphaned(void)
266 read_lock(&tasklist_lock
);
267 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
268 read_unlock(&tasklist_lock
);
273 static bool has_stopped_jobs(struct pid
*pgrp
)
275 struct task_struct
*p
;
277 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
278 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
280 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
286 * Check to see if any process groups have become orphaned as
287 * a result of our exiting, and if they have any stopped jobs,
288 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
291 kill_orphaned_pgrp(struct task_struct
*tsk
, struct task_struct
*parent
)
293 struct pid
*pgrp
= task_pgrp(tsk
);
294 struct task_struct
*ignored_task
= tsk
;
297 /* exit: our father is in a different pgrp than
298 * we are and we were the only connection outside.
300 parent
= tsk
->real_parent
;
302 /* reparent: our child is in a different pgrp than
303 * we are, and it was the only connection outside.
307 if (task_pgrp(parent
) != pgrp
&&
308 task_session(parent
) == task_session(tsk
) &&
309 will_become_orphaned_pgrp(pgrp
, ignored_task
) &&
310 has_stopped_jobs(pgrp
)) {
311 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
312 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
318 * A task is exiting. If it owned this mm, find a new owner for the mm.
320 void mm_update_next_owner(struct mm_struct
*mm
)
322 struct task_struct
*c
, *g
, *p
= current
;
326 * If the exiting or execing task is not the owner, it's
327 * someone else's problem.
332 * The current owner is exiting/execing and there are no other
333 * candidates. Do not leave the mm pointing to a possibly
334 * freed task structure.
336 if (atomic_read(&mm
->mm_users
) <= 1) {
341 read_lock(&tasklist_lock
);
343 * Search in the children
345 list_for_each_entry(c
, &p
->children
, sibling
) {
347 goto assign_new_owner
;
351 * Search in the siblings
353 list_for_each_entry(c
, &p
->real_parent
->children
, sibling
) {
355 goto assign_new_owner
;
359 * Search through everything else, we should not get here often.
361 for_each_process(g
) {
362 if (g
->flags
& PF_KTHREAD
)
364 for_each_thread(g
, c
) {
366 goto assign_new_owner
;
371 read_unlock(&tasklist_lock
);
373 * We found no owner yet mm_users > 1: this implies that we are
374 * most likely racing with swapoff (try_to_unuse()) or /proc or
375 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
384 * The task_lock protects c->mm from changing.
385 * We always want mm->owner->mm == mm
389 * Delay read_unlock() till we have the task_lock()
390 * to ensure that c does not slip away underneath us
392 read_unlock(&tasklist_lock
);
402 #endif /* CONFIG_MEMCG */
405 * Turn us into a lazy TLB process if we
408 static void exit_mm(struct task_struct
*tsk
)
410 struct mm_struct
*mm
= tsk
->mm
;
411 struct core_state
*core_state
;
418 * Serialize with any possible pending coredump.
419 * We must hold mmap_sem around checking core_state
420 * and clearing tsk->mm. The core-inducing thread
421 * will increment ->nr_threads for each thread in the
422 * group with ->mm != NULL.
424 down_read(&mm
->mmap_sem
);
425 core_state
= mm
->core_state
;
427 struct core_thread self
;
429 up_read(&mm
->mmap_sem
);
432 self
.next
= xchg(&core_state
->dumper
.next
, &self
);
434 * Implies mb(), the result of xchg() must be visible
435 * to core_state->dumper.
437 if (atomic_dec_and_test(&core_state
->nr_threads
))
438 complete(&core_state
->startup
);
441 set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
442 if (!self
.task
) /* see coredump_finish() */
444 freezable_schedule();
446 __set_task_state(tsk
, TASK_RUNNING
);
447 down_read(&mm
->mmap_sem
);
449 atomic_inc(&mm
->mm_count
);
450 BUG_ON(mm
!= tsk
->active_mm
);
451 /* more a memory barrier than a real lock */
454 up_read(&mm
->mmap_sem
);
455 enter_lazy_tlb(mm
, current
);
457 mm_update_next_owner(mm
);
459 clear_thread_flag(TIF_MEMDIE
);
462 static struct task_struct
*find_alive_thread(struct task_struct
*p
)
464 struct task_struct
*t
;
466 for_each_thread(p
, t
) {
467 if (!(t
->flags
& PF_EXITING
))
473 static struct task_struct
*find_child_reaper(struct task_struct
*father
)
474 __releases(&tasklist_lock
)
475 __acquires(&tasklist_lock
)
477 struct pid_namespace
*pid_ns
= task_active_pid_ns(father
);
478 struct task_struct
*reaper
= pid_ns
->child_reaper
;
480 if (likely(reaper
!= father
))
483 reaper
= find_alive_thread(father
);
485 pid_ns
->child_reaper
= reaper
;
489 write_unlock_irq(&tasklist_lock
);
490 if (unlikely(pid_ns
== &init_pid_ns
)) {
491 panic("Attempted to kill init! exitcode=0x%08x\n",
492 father
->signal
->group_exit_code
?: father
->exit_code
);
494 zap_pid_ns_processes(pid_ns
);
495 write_lock_irq(&tasklist_lock
);
501 * When we die, we re-parent all our children, and try to:
502 * 1. give them to another thread in our thread group, if such a member exists
503 * 2. give it to the first ancestor process which prctl'd itself as a
504 * child_subreaper for its children (like a service manager)
505 * 3. give it to the init process (PID 1) in our pid namespace
507 static struct task_struct
*find_new_reaper(struct task_struct
*father
,
508 struct task_struct
*child_reaper
)
510 struct task_struct
*thread
, *reaper
;
512 thread
= find_alive_thread(father
);
516 if (father
->signal
->has_child_subreaper
) {
518 * Find the first ->is_child_subreaper ancestor in our pid_ns.
519 * We start from father to ensure we can not look into another
520 * namespace, this is safe because all its threads are dead.
522 for (reaper
= father
;
523 !same_thread_group(reaper
, child_reaper
);
524 reaper
= reaper
->real_parent
) {
525 /* call_usermodehelper() descendants need this check */
526 if (reaper
== &init_task
)
528 if (!reaper
->signal
->is_child_subreaper
)
530 thread
= find_alive_thread(reaper
);
540 * Any that need to be release_task'd are put on the @dead list.
542 static void reparent_leader(struct task_struct
*father
, struct task_struct
*p
,
543 struct list_head
*dead
)
545 if (unlikely(p
->exit_state
== EXIT_DEAD
))
548 /* We don't want people slaying init. */
549 p
->exit_signal
= SIGCHLD
;
551 /* If it has exited notify the new parent about this child's death. */
553 p
->exit_state
== EXIT_ZOMBIE
&& thread_group_empty(p
)) {
554 if (do_notify_parent(p
, p
->exit_signal
)) {
555 p
->exit_state
= EXIT_DEAD
;
556 list_add(&p
->ptrace_entry
, dead
);
560 kill_orphaned_pgrp(p
, father
);
564 * This does two things:
566 * A. Make init inherit all the child processes
567 * B. Check to see if any process groups have become orphaned
568 * as a result of our exiting, and if they have any stopped
569 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
571 static void forget_original_parent(struct task_struct
*father
,
572 struct list_head
*dead
)
574 struct task_struct
*p
, *t
, *reaper
;
576 if (unlikely(!list_empty(&father
->ptraced
)))
577 exit_ptrace(father
, dead
);
579 /* Can drop and reacquire tasklist_lock */
580 reaper
= find_child_reaper(father
);
581 if (list_empty(&father
->children
))
584 reaper
= find_new_reaper(father
, reaper
);
585 list_for_each_entry(p
, &father
->children
, sibling
) {
586 for_each_thread(p
, t
) {
587 t
->real_parent
= reaper
;
588 BUG_ON((!t
->ptrace
) != (t
->parent
== father
));
589 if (likely(!t
->ptrace
))
590 t
->parent
= t
->real_parent
;
591 if (t
->pdeath_signal
)
592 group_send_sig_info(t
->pdeath_signal
,
596 * If this is a threaded reparent there is no need to
597 * notify anyone anything has happened.
599 if (!same_thread_group(reaper
, father
))
600 reparent_leader(father
, p
, dead
);
602 list_splice_tail_init(&father
->children
, &reaper
->children
);
606 * Send signals to all our closest relatives so that they know
607 * to properly mourn us..
609 static void exit_notify(struct task_struct
*tsk
, int group_dead
)
612 struct task_struct
*p
, *n
;
615 write_lock_irq(&tasklist_lock
);
616 forget_original_parent(tsk
, &dead
);
619 kill_orphaned_pgrp(tsk
->group_leader
, NULL
);
621 if (unlikely(tsk
->ptrace
)) {
622 int sig
= thread_group_leader(tsk
) &&
623 thread_group_empty(tsk
) &&
624 !ptrace_reparented(tsk
) ?
625 tsk
->exit_signal
: SIGCHLD
;
626 autoreap
= do_notify_parent(tsk
, sig
);
627 } else if (thread_group_leader(tsk
)) {
628 autoreap
= thread_group_empty(tsk
) &&
629 do_notify_parent(tsk
, tsk
->exit_signal
);
634 tsk
->exit_state
= autoreap
? EXIT_DEAD
: EXIT_ZOMBIE
;
635 if (tsk
->exit_state
== EXIT_DEAD
)
636 list_add(&tsk
->ptrace_entry
, &dead
);
638 /* mt-exec, de_thread() is waiting for group leader */
639 if (unlikely(tsk
->signal
->notify_count
< 0))
640 wake_up_process(tsk
->signal
->group_exit_task
);
641 write_unlock_irq(&tasklist_lock
);
643 list_for_each_entry_safe(p
, n
, &dead
, ptrace_entry
) {
644 list_del_init(&p
->ptrace_entry
);
649 #ifdef CONFIG_DEBUG_STACK_USAGE
650 static void check_stack_usage(void)
652 static DEFINE_SPINLOCK(low_water_lock
);
653 static int lowest_to_date
= THREAD_SIZE
;
656 free
= stack_not_used(current
);
658 if (free
>= lowest_to_date
)
661 spin_lock(&low_water_lock
);
662 if (free
< lowest_to_date
) {
663 pr_warn("%s (%d) used greatest stack depth: %lu bytes left\n",
664 current
->comm
, task_pid_nr(current
), free
);
665 lowest_to_date
= free
;
667 spin_unlock(&low_water_lock
);
670 static inline void check_stack_usage(void) {}
673 void do_exit(long code
)
675 struct task_struct
*tsk
= current
;
677 TASKS_RCU(int tasks_rcu_i
);
679 profile_task_exit(tsk
);
681 WARN_ON(blk_needs_flush_plug(tsk
));
683 if (unlikely(in_interrupt()))
684 panic("Aiee, killing interrupt handler!");
685 if (unlikely(!tsk
->pid
))
686 panic("Attempted to kill the idle task!");
689 * If do_exit is called because this processes oopsed, it's possible
690 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
691 * continuing. Amongst other possible reasons, this is to prevent
692 * mm_release()->clear_child_tid() from writing to a user-controlled
697 ptrace_event(PTRACE_EVENT_EXIT
, code
);
699 validate_creds_for_do_exit(tsk
);
702 * We're taking recursive faults here in do_exit. Safest is to just
703 * leave this task alone and wait for reboot.
705 if (unlikely(tsk
->flags
& PF_EXITING
)) {
706 pr_alert("Fixing recursive fault but reboot is needed!\n");
708 * We can do this unlocked here. The futex code uses
709 * this flag just to verify whether the pi state
710 * cleanup has been done or not. In the worst case it
711 * loops once more. We pretend that the cleanup was
712 * done as there is no way to return. Either the
713 * OWNER_DIED bit is set by now or we push the blocked
714 * task into the wait for ever nirwana as well.
716 tsk
->flags
|= PF_EXITPIDONE
;
717 set_current_state(TASK_UNINTERRUPTIBLE
);
721 exit_signals(tsk
); /* sets PF_EXITING */
723 * tsk->flags are checked in the futex code to protect against
724 * an exiting task cleaning up the robust pi futexes.
727 raw_spin_unlock_wait(&tsk
->pi_lock
);
729 if (unlikely(in_atomic()))
730 pr_info("note: %s[%d] exited with preempt_count %d\n",
731 current
->comm
, task_pid_nr(current
),
734 acct_update_integrals(tsk
);
735 /* sync mm's RSS info before statistics gathering */
737 sync_mm_rss(tsk
->mm
);
738 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
740 hrtimer_cancel(&tsk
->signal
->real_timer
);
741 exit_itimers(tsk
->signal
);
743 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, tsk
->mm
);
745 acct_collect(code
, group_dead
);
750 tsk
->exit_code
= code
;
751 taskstats_exit(tsk
, group_dead
);
757 trace_sched_process_exit(tsk
);
764 disassociate_ctty(1);
765 exit_task_namespaces(tsk
);
770 * Flush inherited counters to the parent - before the parent
771 * gets woken up by child-exit notifications.
773 * because of cgroup mode, must be called before cgroup_exit()
775 perf_event_exit_task(tsk
);
779 module_put(task_thread_info(tsk
)->exec_domain
->module
);
782 * FIXME: do that only when needed, using sched_exit tracepoint
784 flush_ptrace_hw_breakpoint(tsk
);
786 TASKS_RCU(tasks_rcu_i
= __srcu_read_lock(&tasks_rcu_exit_srcu
));
787 exit_notify(tsk
, group_dead
);
788 proc_exit_connector(tsk
);
791 mpol_put(tsk
->mempolicy
);
792 tsk
->mempolicy
= NULL
;
796 if (unlikely(current
->pi_state_cache
))
797 kfree(current
->pi_state_cache
);
800 * Make sure we are holding no locks:
802 debug_check_no_locks_held();
804 * We can do this unlocked here. The futex code uses this flag
805 * just to verify whether the pi state cleanup has been done
806 * or not. In the worst case it loops once more.
808 tsk
->flags
|= PF_EXITPIDONE
;
811 exit_io_context(tsk
);
813 if (tsk
->splice_pipe
)
814 free_pipe_info(tsk
->splice_pipe
);
816 if (tsk
->task_frag
.page
)
817 put_page(tsk
->task_frag
.page
);
819 validate_creds_for_do_exit(tsk
);
824 __this_cpu_add(dirty_throttle_leaks
, tsk
->nr_dirtied
);
826 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu
, tasks_rcu_i
));
829 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
830 * when the following two conditions become true.
831 * - There is race condition of mmap_sem (It is acquired by
833 * - SMI occurs before setting TASK_RUNINNG.
834 * (or hypervisor of virtual machine switches to other guest)
835 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
837 * To avoid it, we have to wait for releasing tsk->pi_lock which
838 * is held by try_to_wake_up()
841 raw_spin_unlock_wait(&tsk
->pi_lock
);
843 /* causes final put_task_struct in finish_task_switch(). */
844 tsk
->state
= TASK_DEAD
;
845 tsk
->flags
|= PF_NOFREEZE
; /* tell freezer to ignore us */
848 /* Avoid "noreturn function does return". */
850 cpu_relax(); /* For when BUG is null */
852 EXPORT_SYMBOL_GPL(do_exit
);
854 void complete_and_exit(struct completion
*comp
, long code
)
861 EXPORT_SYMBOL(complete_and_exit
);
863 SYSCALL_DEFINE1(exit
, int, error_code
)
865 do_exit((error_code
&0xff)<<8);
869 * Take down every thread in the group. This is called by fatal signals
870 * as well as by sys_exit_group (below).
873 do_group_exit(int exit_code
)
875 struct signal_struct
*sig
= current
->signal
;
877 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
879 if (signal_group_exit(sig
))
880 exit_code
= sig
->group_exit_code
;
881 else if (!thread_group_empty(current
)) {
882 struct sighand_struct
*const sighand
= current
->sighand
;
884 spin_lock_irq(&sighand
->siglock
);
885 if (signal_group_exit(sig
))
886 /* Another thread got here before we took the lock. */
887 exit_code
= sig
->group_exit_code
;
889 sig
->group_exit_code
= exit_code
;
890 sig
->flags
= SIGNAL_GROUP_EXIT
;
891 zap_other_threads(current
);
893 spin_unlock_irq(&sighand
->siglock
);
901 * this kills every thread in the thread group. Note that any externally
902 * wait4()-ing process will get the correct exit code - even if this
903 * thread is not the thread group leader.
905 SYSCALL_DEFINE1(exit_group
, int, error_code
)
907 do_group_exit((error_code
& 0xff) << 8);
913 enum pid_type wo_type
;
917 struct siginfo __user
*wo_info
;
919 struct rusage __user
*wo_rusage
;
921 wait_queue_t child_wait
;
926 struct pid
*task_pid_type(struct task_struct
*task
, enum pid_type type
)
928 if (type
!= PIDTYPE_PID
)
929 task
= task
->group_leader
;
930 return task
->pids
[type
].pid
;
933 static int eligible_pid(struct wait_opts
*wo
, struct task_struct
*p
)
935 return wo
->wo_type
== PIDTYPE_MAX
||
936 task_pid_type(p
, wo
->wo_type
) == wo
->wo_pid
;
939 static int eligible_child(struct wait_opts
*wo
, struct task_struct
*p
)
941 if (!eligible_pid(wo
, p
))
943 /* Wait for all children (clone and not) if __WALL is set;
944 * otherwise, wait for clone children *only* if __WCLONE is
945 * set; otherwise, wait for non-clone children *only*. (Note:
946 * A "clone" child here is one that reports to its parent
947 * using a signal other than SIGCHLD.) */
948 if (((p
->exit_signal
!= SIGCHLD
) ^ !!(wo
->wo_flags
& __WCLONE
))
949 && !(wo
->wo_flags
& __WALL
))
955 static int wait_noreap_copyout(struct wait_opts
*wo
, struct task_struct
*p
,
956 pid_t pid
, uid_t uid
, int why
, int status
)
958 struct siginfo __user
*infop
;
959 int retval
= wo
->wo_rusage
960 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
966 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
968 retval
= put_user(0, &infop
->si_errno
);
970 retval
= put_user((short)why
, &infop
->si_code
);
972 retval
= put_user(pid
, &infop
->si_pid
);
974 retval
= put_user(uid
, &infop
->si_uid
);
976 retval
= put_user(status
, &infop
->si_status
);
984 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
985 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
986 * the lock and this task is uninteresting. If we return nonzero, we have
987 * released the lock and the system call should return.
989 static int wait_task_zombie(struct wait_opts
*wo
, struct task_struct
*p
)
991 int state
, retval
, status
;
992 pid_t pid
= task_pid_vnr(p
);
993 uid_t uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
994 struct siginfo __user
*infop
;
996 if (!likely(wo
->wo_flags
& WEXITED
))
999 if (unlikely(wo
->wo_flags
& WNOWAIT
)) {
1000 int exit_code
= p
->exit_code
;
1004 read_unlock(&tasklist_lock
);
1005 sched_annotate_sleep();
1007 if ((exit_code
& 0x7f) == 0) {
1009 status
= exit_code
>> 8;
1011 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1012 status
= exit_code
& 0x7f;
1014 return wait_noreap_copyout(wo
, p
, pid
, uid
, why
, status
);
1017 * Move the task's state to DEAD/TRACE, only one thread can do this.
1019 state
= (ptrace_reparented(p
) && thread_group_leader(p
)) ?
1020 EXIT_TRACE
: EXIT_DEAD
;
1021 if (cmpxchg(&p
->exit_state
, EXIT_ZOMBIE
, state
) != EXIT_ZOMBIE
)
1024 * We own this thread, nobody else can reap it.
1026 read_unlock(&tasklist_lock
);
1027 sched_annotate_sleep();
1030 * Check thread_group_leader() to exclude the traced sub-threads.
1032 if (state
== EXIT_DEAD
&& thread_group_leader(p
)) {
1033 struct signal_struct
*sig
= p
->signal
;
1034 struct signal_struct
*psig
= current
->signal
;
1035 unsigned long maxrss
;
1036 cputime_t tgutime
, tgstime
;
1039 * The resource counters for the group leader are in its
1040 * own task_struct. Those for dead threads in the group
1041 * are in its signal_struct, as are those for the child
1042 * processes it has previously reaped. All these
1043 * accumulate in the parent's signal_struct c* fields.
1045 * We don't bother to take a lock here to protect these
1046 * p->signal fields because the whole thread group is dead
1047 * and nobody can change them.
1049 * psig->stats_lock also protects us from our sub-theads
1050 * which can reap other children at the same time. Until
1051 * we change k_getrusage()-like users to rely on this lock
1052 * we have to take ->siglock as well.
1054 * We use thread_group_cputime_adjusted() to get times for
1055 * the thread group, which consolidates times for all threads
1056 * in the group including the group leader.
1058 thread_group_cputime_adjusted(p
, &tgutime
, &tgstime
);
1059 spin_lock_irq(¤t
->sighand
->siglock
);
1060 write_seqlock(&psig
->stats_lock
);
1061 psig
->cutime
+= tgutime
+ sig
->cutime
;
1062 psig
->cstime
+= tgstime
+ sig
->cstime
;
1063 psig
->cgtime
+= task_gtime(p
) + sig
->gtime
+ sig
->cgtime
;
1065 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1067 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1069 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1071 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1073 task_io_get_inblock(p
) +
1074 sig
->inblock
+ sig
->cinblock
;
1076 task_io_get_oublock(p
) +
1077 sig
->oublock
+ sig
->coublock
;
1078 maxrss
= max(sig
->maxrss
, sig
->cmaxrss
);
1079 if (psig
->cmaxrss
< maxrss
)
1080 psig
->cmaxrss
= maxrss
;
1081 task_io_accounting_add(&psig
->ioac
, &p
->ioac
);
1082 task_io_accounting_add(&psig
->ioac
, &sig
->ioac
);
1083 write_sequnlock(&psig
->stats_lock
);
1084 spin_unlock_irq(¤t
->sighand
->siglock
);
1087 retval
= wo
->wo_rusage
1088 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1089 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1090 ? p
->signal
->group_exit_code
: p
->exit_code
;
1091 if (!retval
&& wo
->wo_stat
)
1092 retval
= put_user(status
, wo
->wo_stat
);
1094 infop
= wo
->wo_info
;
1095 if (!retval
&& infop
)
1096 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1097 if (!retval
&& infop
)
1098 retval
= put_user(0, &infop
->si_errno
);
1099 if (!retval
&& infop
) {
1102 if ((status
& 0x7f) == 0) {
1106 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1109 retval
= put_user((short)why
, &infop
->si_code
);
1111 retval
= put_user(status
, &infop
->si_status
);
1113 if (!retval
&& infop
)
1114 retval
= put_user(pid
, &infop
->si_pid
);
1115 if (!retval
&& infop
)
1116 retval
= put_user(uid
, &infop
->si_uid
);
1120 if (state
== EXIT_TRACE
) {
1121 write_lock_irq(&tasklist_lock
);
1122 /* We dropped tasklist, ptracer could die and untrace */
1125 /* If parent wants a zombie, don't release it now */
1126 state
= EXIT_ZOMBIE
;
1127 if (do_notify_parent(p
, p
->exit_signal
))
1129 p
->exit_state
= state
;
1130 write_unlock_irq(&tasklist_lock
);
1132 if (state
== EXIT_DEAD
)
1138 static int *task_stopped_code(struct task_struct
*p
, bool ptrace
)
1141 if (task_is_stopped_or_traced(p
) &&
1142 !(p
->jobctl
& JOBCTL_LISTENING
))
1143 return &p
->exit_code
;
1145 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
1146 return &p
->signal
->group_exit_code
;
1152 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1154 * @ptrace: is the wait for ptrace
1155 * @p: task to wait for
1157 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1160 * read_lock(&tasklist_lock), which is released if return value is
1161 * non-zero. Also, grabs and releases @p->sighand->siglock.
1164 * 0 if wait condition didn't exist and search for other wait conditions
1165 * should continue. Non-zero return, -errno on failure and @p's pid on
1166 * success, implies that tasklist_lock is released and wait condition
1167 * search should terminate.
1169 static int wait_task_stopped(struct wait_opts
*wo
,
1170 int ptrace
, struct task_struct
*p
)
1172 struct siginfo __user
*infop
;
1173 int retval
, exit_code
, *p_code
, why
;
1174 uid_t uid
= 0; /* unneeded, required by compiler */
1178 * Traditionally we see ptrace'd stopped tasks regardless of options.
1180 if (!ptrace
&& !(wo
->wo_flags
& WUNTRACED
))
1183 if (!task_stopped_code(p
, ptrace
))
1187 spin_lock_irq(&p
->sighand
->siglock
);
1189 p_code
= task_stopped_code(p
, ptrace
);
1190 if (unlikely(!p_code
))
1193 exit_code
= *p_code
;
1197 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1200 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1202 spin_unlock_irq(&p
->sighand
->siglock
);
1207 * Now we are pretty sure this task is interesting.
1208 * Make sure it doesn't get reaped out from under us while we
1209 * give up the lock and then examine it below. We don't want to
1210 * keep holding onto the tasklist_lock while we call getrusage and
1211 * possibly take page faults for user memory.
1214 pid
= task_pid_vnr(p
);
1215 why
= ptrace
? CLD_TRAPPED
: CLD_STOPPED
;
1216 read_unlock(&tasklist_lock
);
1217 sched_annotate_sleep();
1219 if (unlikely(wo
->wo_flags
& WNOWAIT
))
1220 return wait_noreap_copyout(wo
, p
, pid
, uid
, why
, exit_code
);
1222 retval
= wo
->wo_rusage
1223 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1224 if (!retval
&& wo
->wo_stat
)
1225 retval
= put_user((exit_code
<< 8) | 0x7f, wo
->wo_stat
);
1227 infop
= wo
->wo_info
;
1228 if (!retval
&& infop
)
1229 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1230 if (!retval
&& infop
)
1231 retval
= put_user(0, &infop
->si_errno
);
1232 if (!retval
&& infop
)
1233 retval
= put_user((short)why
, &infop
->si_code
);
1234 if (!retval
&& infop
)
1235 retval
= put_user(exit_code
, &infop
->si_status
);
1236 if (!retval
&& infop
)
1237 retval
= put_user(pid
, &infop
->si_pid
);
1238 if (!retval
&& infop
)
1239 retval
= put_user(uid
, &infop
->si_uid
);
1249 * Handle do_wait work for one task in a live, non-stopped state.
1250 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1251 * the lock and this task is uninteresting. If we return nonzero, we have
1252 * released the lock and the system call should return.
1254 static int wait_task_continued(struct wait_opts
*wo
, struct task_struct
*p
)
1260 if (!unlikely(wo
->wo_flags
& WCONTINUED
))
1263 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1266 spin_lock_irq(&p
->sighand
->siglock
);
1267 /* Re-check with the lock held. */
1268 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1269 spin_unlock_irq(&p
->sighand
->siglock
);
1272 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1273 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1274 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1275 spin_unlock_irq(&p
->sighand
->siglock
);
1277 pid
= task_pid_vnr(p
);
1279 read_unlock(&tasklist_lock
);
1280 sched_annotate_sleep();
1283 retval
= wo
->wo_rusage
1284 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1286 if (!retval
&& wo
->wo_stat
)
1287 retval
= put_user(0xffff, wo
->wo_stat
);
1291 retval
= wait_noreap_copyout(wo
, p
, pid
, uid
,
1292 CLD_CONTINUED
, SIGCONT
);
1293 BUG_ON(retval
== 0);
1300 * Consider @p for a wait by @parent.
1302 * -ECHILD should be in ->notask_error before the first call.
1303 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1304 * Returns zero if the search for a child should continue;
1305 * then ->notask_error is 0 if @p is an eligible child,
1306 * or another error from security_task_wait(), or still -ECHILD.
1308 static int wait_consider_task(struct wait_opts
*wo
, int ptrace
,
1309 struct task_struct
*p
)
1313 if (unlikely(p
->exit_state
== EXIT_DEAD
))
1316 ret
= eligible_child(wo
, p
);
1320 ret
= security_task_wait(p
);
1321 if (unlikely(ret
< 0)) {
1323 * If we have not yet seen any eligible child,
1324 * then let this error code replace -ECHILD.
1325 * A permission error will give the user a clue
1326 * to look for security policy problems, rather
1327 * than for mysterious wait bugs.
1329 if (wo
->notask_error
)
1330 wo
->notask_error
= ret
;
1334 if (unlikely(p
->exit_state
== EXIT_TRACE
)) {
1336 * ptrace == 0 means we are the natural parent. In this case
1337 * we should clear notask_error, debugger will notify us.
1339 if (likely(!ptrace
))
1340 wo
->notask_error
= 0;
1344 if (likely(!ptrace
) && unlikely(p
->ptrace
)) {
1346 * If it is traced by its real parent's group, just pretend
1347 * the caller is ptrace_do_wait() and reap this child if it
1350 * This also hides group stop state from real parent; otherwise
1351 * a single stop can be reported twice as group and ptrace stop.
1352 * If a ptracer wants to distinguish these two events for its
1353 * own children it should create a separate process which takes
1354 * the role of real parent.
1356 if (!ptrace_reparented(p
))
1361 if (p
->exit_state
== EXIT_ZOMBIE
) {
1362 /* we don't reap group leaders with subthreads */
1363 if (!delay_group_leader(p
)) {
1365 * A zombie ptracee is only visible to its ptracer.
1366 * Notification and reaping will be cascaded to the
1367 * real parent when the ptracer detaches.
1369 if (unlikely(ptrace
) || likely(!p
->ptrace
))
1370 return wait_task_zombie(wo
, p
);
1374 * Allow access to stopped/continued state via zombie by
1375 * falling through. Clearing of notask_error is complex.
1379 * If WEXITED is set, notask_error should naturally be
1380 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1381 * so, if there are live subthreads, there are events to
1382 * wait for. If all subthreads are dead, it's still safe
1383 * to clear - this function will be called again in finite
1384 * amount time once all the subthreads are released and
1385 * will then return without clearing.
1389 * Stopped state is per-task and thus can't change once the
1390 * target task dies. Only continued and exited can happen.
1391 * Clear notask_error if WCONTINUED | WEXITED.
1393 if (likely(!ptrace
) || (wo
->wo_flags
& (WCONTINUED
| WEXITED
)))
1394 wo
->notask_error
= 0;
1397 * @p is alive and it's gonna stop, continue or exit, so
1398 * there always is something to wait for.
1400 wo
->notask_error
= 0;
1404 * Wait for stopped. Depending on @ptrace, different stopped state
1405 * is used and the two don't interact with each other.
1407 ret
= wait_task_stopped(wo
, ptrace
, p
);
1412 * Wait for continued. There's only one continued state and the
1413 * ptracer can consume it which can confuse the real parent. Don't
1414 * use WCONTINUED from ptracer. You don't need or want it.
1416 return wait_task_continued(wo
, p
);
1420 * Do the work of do_wait() for one thread in the group, @tsk.
1422 * -ECHILD should be in ->notask_error before the first call.
1423 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1424 * Returns zero if the search for a child should continue; then
1425 * ->notask_error is 0 if there were any eligible children,
1426 * or another error from security_task_wait(), or still -ECHILD.
1428 static int do_wait_thread(struct wait_opts
*wo
, struct task_struct
*tsk
)
1430 struct task_struct
*p
;
1432 list_for_each_entry(p
, &tsk
->children
, sibling
) {
1433 int ret
= wait_consider_task(wo
, 0, p
);
1442 static int ptrace_do_wait(struct wait_opts
*wo
, struct task_struct
*tsk
)
1444 struct task_struct
*p
;
1446 list_for_each_entry(p
, &tsk
->ptraced
, ptrace_entry
) {
1447 int ret
= wait_consider_task(wo
, 1, p
);
1456 static int child_wait_callback(wait_queue_t
*wait
, unsigned mode
,
1457 int sync
, void *key
)
1459 struct wait_opts
*wo
= container_of(wait
, struct wait_opts
,
1461 struct task_struct
*p
= key
;
1463 if (!eligible_pid(wo
, p
))
1466 if ((wo
->wo_flags
& __WNOTHREAD
) && wait
->private != p
->parent
)
1469 return default_wake_function(wait
, mode
, sync
, key
);
1472 void __wake_up_parent(struct task_struct
*p
, struct task_struct
*parent
)
1474 __wake_up_sync_key(&parent
->signal
->wait_chldexit
,
1475 TASK_INTERRUPTIBLE
, 1, p
);
1478 static long do_wait(struct wait_opts
*wo
)
1480 struct task_struct
*tsk
;
1483 trace_sched_process_wait(wo
->wo_pid
);
1485 init_waitqueue_func_entry(&wo
->child_wait
, child_wait_callback
);
1486 wo
->child_wait
.private = current
;
1487 add_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1490 * If there is nothing that can match our critiera just get out.
1491 * We will clear ->notask_error to zero if we see any child that
1492 * might later match our criteria, even if we are not able to reap
1495 wo
->notask_error
= -ECHILD
;
1496 if ((wo
->wo_type
< PIDTYPE_MAX
) &&
1497 (!wo
->wo_pid
|| hlist_empty(&wo
->wo_pid
->tasks
[wo
->wo_type
])))
1500 set_current_state(TASK_INTERRUPTIBLE
);
1501 read_lock(&tasklist_lock
);
1504 retval
= do_wait_thread(wo
, tsk
);
1508 retval
= ptrace_do_wait(wo
, tsk
);
1512 if (wo
->wo_flags
& __WNOTHREAD
)
1514 } while_each_thread(current
, tsk
);
1515 read_unlock(&tasklist_lock
);
1518 retval
= wo
->notask_error
;
1519 if (!retval
&& !(wo
->wo_flags
& WNOHANG
)) {
1520 retval
= -ERESTARTSYS
;
1521 if (!signal_pending(current
)) {
1527 __set_current_state(TASK_RUNNING
);
1528 remove_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1532 SYSCALL_DEFINE5(waitid
, int, which
, pid_t
, upid
, struct siginfo __user
*,
1533 infop
, int, options
, struct rusage __user
*, ru
)
1535 struct wait_opts wo
;
1536 struct pid
*pid
= NULL
;
1540 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
))
1542 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1555 type
= PIDTYPE_PGID
;
1563 if (type
< PIDTYPE_MAX
)
1564 pid
= find_get_pid(upid
);
1568 wo
.wo_flags
= options
;
1578 * For a WNOHANG return, clear out all the fields
1579 * we would set so the user can easily tell the
1583 ret
= put_user(0, &infop
->si_signo
);
1585 ret
= put_user(0, &infop
->si_errno
);
1587 ret
= put_user(0, &infop
->si_code
);
1589 ret
= put_user(0, &infop
->si_pid
);
1591 ret
= put_user(0, &infop
->si_uid
);
1593 ret
= put_user(0, &infop
->si_status
);
1600 SYSCALL_DEFINE4(wait4
, pid_t
, upid
, int __user
*, stat_addr
,
1601 int, options
, struct rusage __user
*, ru
)
1603 struct wait_opts wo
;
1604 struct pid
*pid
= NULL
;
1608 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1609 __WNOTHREAD
|__WCLONE
|__WALL
))
1614 else if (upid
< 0) {
1615 type
= PIDTYPE_PGID
;
1616 pid
= find_get_pid(-upid
);
1617 } else if (upid
== 0) {
1618 type
= PIDTYPE_PGID
;
1619 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1620 } else /* upid > 0 */ {
1622 pid
= find_get_pid(upid
);
1627 wo
.wo_flags
= options
| WEXITED
;
1629 wo
.wo_stat
= stat_addr
;
1637 #ifdef __ARCH_WANT_SYS_WAITPID
1640 * sys_waitpid() remains for compatibility. waitpid() should be
1641 * implemented by calling sys_wait4() from libc.a.
1643 SYSCALL_DEFINE3(waitpid
, pid_t
, pid
, int __user
*, stat_addr
, int, options
)
1645 return sys_wait4(pid
, stat_addr
, options
, NULL
);