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
);
119 * Accumulate here the counters for all threads but the
120 * group leader as they die, so they can be added into
121 * the process-wide totals when those are taken.
122 * The group leader stays around as a zombie as long
123 * as there are other threads. When it gets reaped,
124 * the exit.c code will add its counts into these totals.
125 * We won't ever get here for the group leader, since it
126 * will have been the last reference on the signal_struct.
128 task_cputime(tsk
, &utime
, &stime
);
131 sig
->gtime
+= task_gtime(tsk
);
132 sig
->min_flt
+= tsk
->min_flt
;
133 sig
->maj_flt
+= tsk
->maj_flt
;
134 sig
->nvcsw
+= tsk
->nvcsw
;
135 sig
->nivcsw
+= tsk
->nivcsw
;
136 sig
->inblock
+= task_io_get_inblock(tsk
);
137 sig
->oublock
+= task_io_get_oublock(tsk
);
138 task_io_accounting_add(&sig
->ioac
, &tsk
->ioac
);
139 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
143 __unhash_process(tsk
, group_dead
);
146 * Do this under ->siglock, we can race with another thread
147 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
149 flush_sigqueue(&tsk
->pending
);
151 spin_unlock(&sighand
->siglock
);
153 __cleanup_sighand(sighand
);
154 clear_tsk_thread_flag(tsk
,TIF_SIGPENDING
);
156 flush_sigqueue(&sig
->shared_pending
);
161 static void delayed_put_task_struct(struct rcu_head
*rhp
)
163 struct task_struct
*tsk
= container_of(rhp
, struct task_struct
, rcu
);
165 perf_event_delayed_put(tsk
);
166 trace_sched_process_free(tsk
);
167 put_task_struct(tsk
);
171 void release_task(struct task_struct
* p
)
173 struct task_struct
*leader
;
176 /* don't need to get the RCU readlock here - the process is dead and
177 * can't be modifying its own credentials. But shut RCU-lockdep up */
179 atomic_dec(&__task_cred(p
)->user
->processes
);
184 write_lock_irq(&tasklist_lock
);
185 ptrace_release_task(p
);
189 * If we are the last non-leader member of the thread
190 * group, and the leader is zombie, then notify the
191 * group leader's parent process. (if it wants notification.)
194 leader
= p
->group_leader
;
195 if (leader
!= p
&& thread_group_empty(leader
) && leader
->exit_state
== EXIT_ZOMBIE
) {
197 * If we were the last child thread and the leader has
198 * exited already, and the leader's parent ignores SIGCHLD,
199 * then we are the one who should release the leader.
201 zap_leader
= do_notify_parent(leader
, leader
->exit_signal
);
203 leader
->exit_state
= EXIT_DEAD
;
206 write_unlock_irq(&tasklist_lock
);
208 call_rcu(&p
->rcu
, delayed_put_task_struct
);
211 if (unlikely(zap_leader
))
216 * This checks not only the pgrp, but falls back on the pid if no
217 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
220 * The caller must hold rcu lock or the tasklist lock.
222 struct pid
*session_of_pgrp(struct pid
*pgrp
)
224 struct task_struct
*p
;
225 struct pid
*sid
= NULL
;
227 p
= pid_task(pgrp
, PIDTYPE_PGID
);
229 p
= pid_task(pgrp
, PIDTYPE_PID
);
231 sid
= task_session(p
);
237 * Determine if a process group is "orphaned", according to the POSIX
238 * definition in 2.2.2.52. Orphaned process groups are not to be affected
239 * by terminal-generated stop signals. Newly orphaned process groups are
240 * to receive a SIGHUP and a SIGCONT.
242 * "I ask you, have you ever known what it is to be an orphan?"
244 static int will_become_orphaned_pgrp(struct pid
*pgrp
, 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
);
316 void __set_special_pids(struct pid
*pid
)
318 struct task_struct
*curr
= current
->group_leader
;
320 if (task_session(curr
) != pid
)
321 change_pid(curr
, PIDTYPE_SID
, pid
);
323 if (task_pgrp(curr
) != pid
)
324 change_pid(curr
, PIDTYPE_PGID
, pid
);
328 * Let kernel threads use this to say that they allow a certain signal.
329 * Must not be used if kthread was cloned with CLONE_SIGHAND.
331 int allow_signal(int sig
)
333 if (!valid_signal(sig
) || sig
< 1)
336 spin_lock_irq(¤t
->sighand
->siglock
);
337 /* This is only needed for daemonize()'ed kthreads */
338 sigdelset(¤t
->blocked
, sig
);
340 * Kernel threads handle their own signals. Let the signal code
341 * know it'll be handled, so that they don't get converted to
342 * SIGKILL or just silently dropped.
344 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= (void __user
*)2;
346 spin_unlock_irq(¤t
->sighand
->siglock
);
350 EXPORT_SYMBOL(allow_signal
);
352 int disallow_signal(int sig
)
354 if (!valid_signal(sig
) || sig
< 1)
357 spin_lock_irq(¤t
->sighand
->siglock
);
358 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= SIG_IGN
;
360 spin_unlock_irq(¤t
->sighand
->siglock
);
364 EXPORT_SYMBOL(disallow_signal
);
366 #ifdef CONFIG_MM_OWNER
368 * A task is exiting. If it owned this mm, find a new owner for the mm.
370 void mm_update_next_owner(struct mm_struct
*mm
)
372 struct task_struct
*c
, *g
, *p
= current
;
376 * If the exiting or execing task is not the owner, it's
377 * someone else's problem.
382 * The current owner is exiting/execing and there are no other
383 * candidates. Do not leave the mm pointing to a possibly
384 * freed task structure.
386 if (atomic_read(&mm
->mm_users
) <= 1) {
391 read_lock(&tasklist_lock
);
393 * Search in the children
395 list_for_each_entry(c
, &p
->children
, sibling
) {
397 goto assign_new_owner
;
401 * Search in the siblings
403 list_for_each_entry(c
, &p
->real_parent
->children
, sibling
) {
405 goto assign_new_owner
;
409 * Search through everything else. We should not get
412 do_each_thread(g
, c
) {
414 goto assign_new_owner
;
415 } while_each_thread(g
, c
);
417 read_unlock(&tasklist_lock
);
419 * We found no owner yet mm_users > 1: this implies that we are
420 * most likely racing with swapoff (try_to_unuse()) or /proc or
421 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
430 * The task_lock protects c->mm from changing.
431 * We always want mm->owner->mm == mm
435 * Delay read_unlock() till we have the task_lock()
436 * to ensure that c does not slip away underneath us
438 read_unlock(&tasklist_lock
);
448 #endif /* CONFIG_MM_OWNER */
451 * Turn us into a lazy TLB process if we
454 static void exit_mm(struct task_struct
* tsk
)
456 struct mm_struct
*mm
= tsk
->mm
;
457 struct core_state
*core_state
;
464 * Serialize with any possible pending coredump.
465 * We must hold mmap_sem around checking core_state
466 * and clearing tsk->mm. The core-inducing thread
467 * will increment ->nr_threads for each thread in the
468 * group with ->mm != NULL.
470 down_read(&mm
->mmap_sem
);
471 core_state
= mm
->core_state
;
473 struct core_thread self
;
474 up_read(&mm
->mmap_sem
);
477 self
.next
= xchg(&core_state
->dumper
.next
, &self
);
479 * Implies mb(), the result of xchg() must be visible
480 * to core_state->dumper.
482 if (atomic_dec_and_test(&core_state
->nr_threads
))
483 complete(&core_state
->startup
);
486 set_task_state(tsk
, TASK_UNINTERRUPTIBLE
);
487 if (!self
.task
) /* see coredump_finish() */
489 freezable_schedule();
491 __set_task_state(tsk
, TASK_RUNNING
);
492 down_read(&mm
->mmap_sem
);
494 atomic_inc(&mm
->mm_count
);
495 BUG_ON(mm
!= tsk
->active_mm
);
496 /* more a memory barrier than a real lock */
499 up_read(&mm
->mmap_sem
);
500 enter_lazy_tlb(mm
, current
);
502 mm_update_next_owner(mm
);
507 * When we die, we re-parent all our children, and try to:
508 * 1. give them to another thread in our thread group, if such a member exists
509 * 2. give it to the first ancestor process which prctl'd itself as a
510 * child_subreaper for its children (like a service manager)
511 * 3. give it to the init process (PID 1) in our pid namespace
513 static struct task_struct
*find_new_reaper(struct task_struct
*father
)
514 __releases(&tasklist_lock
)
515 __acquires(&tasklist_lock
)
517 struct pid_namespace
*pid_ns
= task_active_pid_ns(father
);
518 struct task_struct
*thread
;
521 while_each_thread(father
, thread
) {
522 if (thread
->flags
& PF_EXITING
)
524 if (unlikely(pid_ns
->child_reaper
== father
))
525 pid_ns
->child_reaper
= thread
;
529 if (unlikely(pid_ns
->child_reaper
== father
)) {
530 write_unlock_irq(&tasklist_lock
);
531 if (unlikely(pid_ns
== &init_pid_ns
)) {
532 panic("Attempted to kill init! exitcode=0x%08x\n",
533 father
->signal
->group_exit_code
?:
537 zap_pid_ns_processes(pid_ns
);
538 write_lock_irq(&tasklist_lock
);
539 } else if (father
->signal
->has_child_subreaper
) {
540 struct task_struct
*reaper
;
543 * Find the first ancestor marked as child_subreaper.
544 * Note that the code below checks same_thread_group(reaper,
545 * pid_ns->child_reaper). This is what we need to DTRT in a
546 * PID namespace. However we still need the check above, see
547 * http://marc.info/?l=linux-kernel&m=131385460420380
549 for (reaper
= father
->real_parent
;
550 reaper
!= &init_task
;
551 reaper
= reaper
->real_parent
) {
552 if (same_thread_group(reaper
, pid_ns
->child_reaper
))
554 if (!reaper
->signal
->is_child_subreaper
)
558 if (!(thread
->flags
& PF_EXITING
))
560 } while_each_thread(reaper
, thread
);
564 return pid_ns
->child_reaper
;
568 * Any that need to be release_task'd are put on the @dead list.
570 static void reparent_leader(struct task_struct
*father
, struct task_struct
*p
,
571 struct list_head
*dead
)
573 list_move_tail(&p
->sibling
, &p
->real_parent
->children
);
575 * If this is a threaded reparent there is no need to
576 * notify anyone anything has happened.
578 if (same_thread_group(p
->real_parent
, father
))
582 * We don't want people slaying init.
584 * Note: we do this even if it is EXIT_DEAD, wait_task_zombie()
585 * can change ->exit_state to EXIT_ZOMBIE. If this is the final
586 * state, do_notify_parent() was already called and ->exit_signal
589 p
->exit_signal
= SIGCHLD
;
591 if (p
->exit_state
== EXIT_DEAD
)
594 /* If it has exited notify the new parent about this child's death. */
596 p
->exit_state
== EXIT_ZOMBIE
&& thread_group_empty(p
)) {
597 if (do_notify_parent(p
, p
->exit_signal
)) {
598 p
->exit_state
= EXIT_DEAD
;
599 list_move_tail(&p
->sibling
, dead
);
603 kill_orphaned_pgrp(p
, father
);
606 static void forget_original_parent(struct task_struct
*father
)
608 struct task_struct
*p
, *n
, *reaper
;
609 LIST_HEAD(dead_children
);
611 write_lock_irq(&tasklist_lock
);
613 * Note that exit_ptrace() and find_new_reaper() might
614 * drop tasklist_lock and reacquire it.
617 reaper
= find_new_reaper(father
);
619 list_for_each_entry_safe(p
, n
, &father
->children
, sibling
) {
620 struct task_struct
*t
= p
;
622 t
->real_parent
= reaper
;
623 if (t
->parent
== father
) {
625 t
->parent
= t
->real_parent
;
627 if (t
->pdeath_signal
)
628 group_send_sig_info(t
->pdeath_signal
,
630 } while_each_thread(p
, t
);
631 reparent_leader(father
, p
, &dead_children
);
633 write_unlock_irq(&tasklist_lock
);
635 BUG_ON(!list_empty(&father
->children
));
637 list_for_each_entry_safe(p
, n
, &dead_children
, sibling
) {
638 list_del_init(&p
->sibling
);
644 * Send signals to all our closest relatives so that they know
645 * to properly mourn us..
647 static void exit_notify(struct task_struct
*tsk
, int group_dead
)
652 * This does two things:
654 * A. Make init inherit all the child processes
655 * B. Check to see if any process groups have become orphaned
656 * as a result of our exiting, and if they have any stopped
657 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
659 forget_original_parent(tsk
);
661 write_lock_irq(&tasklist_lock
);
663 kill_orphaned_pgrp(tsk
->group_leader
, NULL
);
665 if (unlikely(tsk
->ptrace
)) {
666 int sig
= thread_group_leader(tsk
) &&
667 thread_group_empty(tsk
) &&
668 !ptrace_reparented(tsk
) ?
669 tsk
->exit_signal
: SIGCHLD
;
670 autoreap
= do_notify_parent(tsk
, sig
);
671 } else if (thread_group_leader(tsk
)) {
672 autoreap
= thread_group_empty(tsk
) &&
673 do_notify_parent(tsk
, tsk
->exit_signal
);
678 tsk
->exit_state
= autoreap
? EXIT_DEAD
: EXIT_ZOMBIE
;
680 /* mt-exec, de_thread() is waiting for group leader */
681 if (unlikely(tsk
->signal
->notify_count
< 0))
682 wake_up_process(tsk
->signal
->group_exit_task
);
683 write_unlock_irq(&tasklist_lock
);
685 /* If the process is dead, release it - nobody will wait for it */
690 #ifdef CONFIG_DEBUG_STACK_USAGE
691 static void check_stack_usage(void)
693 static DEFINE_SPINLOCK(low_water_lock
);
694 static int lowest_to_date
= THREAD_SIZE
;
697 free
= stack_not_used(current
);
699 if (free
>= lowest_to_date
)
702 spin_lock(&low_water_lock
);
703 if (free
< lowest_to_date
) {
704 printk(KERN_WARNING
"%s (%d) used greatest stack depth: "
706 current
->comm
, task_pid_nr(current
), free
);
707 lowest_to_date
= free
;
709 spin_unlock(&low_water_lock
);
712 static inline void check_stack_usage(void) {}
715 void do_exit(long code
)
717 struct task_struct
*tsk
= current
;
720 profile_task_exit(tsk
);
722 WARN_ON(blk_needs_flush_plug(tsk
));
724 if (unlikely(in_interrupt()))
725 panic("Aiee, killing interrupt handler!");
726 if (unlikely(!tsk
->pid
))
727 panic("Attempted to kill the idle task!");
730 * If do_exit is called because this processes oopsed, it's possible
731 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
732 * continuing. Amongst other possible reasons, this is to prevent
733 * mm_release()->clear_child_tid() from writing to a user-controlled
738 ptrace_event(PTRACE_EVENT_EXIT
, code
);
740 validate_creds_for_do_exit(tsk
);
743 * We're taking recursive faults here in do_exit. Safest is to just
744 * leave this task alone and wait for reboot.
746 if (unlikely(tsk
->flags
& PF_EXITING
)) {
748 "Fixing recursive fault but reboot is needed!\n");
750 * We can do this unlocked here. The futex code uses
751 * this flag just to verify whether the pi state
752 * cleanup has been done or not. In the worst case it
753 * loops once more. We pretend that the cleanup was
754 * done as there is no way to return. Either the
755 * OWNER_DIED bit is set by now or we push the blocked
756 * task into the wait for ever nirwana as well.
758 tsk
->flags
|= PF_EXITPIDONE
;
759 set_current_state(TASK_UNINTERRUPTIBLE
);
763 exit_signals(tsk
); /* sets PF_EXITING */
765 * tsk->flags are checked in the futex code to protect against
766 * an exiting task cleaning up the robust pi futexes.
769 raw_spin_unlock_wait(&tsk
->pi_lock
);
771 if (unlikely(in_atomic()))
772 printk(KERN_INFO
"note: %s[%d] exited with preempt_count %d\n",
773 current
->comm
, task_pid_nr(current
),
776 acct_update_integrals(tsk
);
777 /* sync mm's RSS info before statistics gathering */
779 sync_mm_rss(tsk
->mm
);
780 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
782 hrtimer_cancel(&tsk
->signal
->real_timer
);
783 exit_itimers(tsk
->signal
);
785 setmax_mm_hiwater_rss(&tsk
->signal
->maxrss
, tsk
->mm
);
787 acct_collect(code
, group_dead
);
792 tsk
->exit_code
= code
;
793 taskstats_exit(tsk
, group_dead
);
799 trace_sched_process_exit(tsk
);
806 disassociate_ctty(1);
807 exit_task_namespaces(tsk
);
813 * Flush inherited counters to the parent - before the parent
814 * gets woken up by child-exit notifications.
816 * because of cgroup mode, must be called before cgroup_exit()
818 perf_event_exit_task(tsk
);
822 module_put(task_thread_info(tsk
)->exec_domain
->module
);
824 proc_exit_connector(tsk
);
826 * FIXME: do that only when needed, using sched_exit tracepoint
828 ptrace_put_breakpoints(tsk
);
830 exit_notify(tsk
, group_dead
);
833 mpol_put(tsk
->mempolicy
);
834 tsk
->mempolicy
= NULL
;
838 if (unlikely(current
->pi_state_cache
))
839 kfree(current
->pi_state_cache
);
842 * Make sure we are holding no locks:
844 debug_check_no_locks_held(tsk
);
846 * We can do this unlocked here. The futex code uses this flag
847 * just to verify whether the pi state cleanup has been done
848 * or not. In the worst case it loops once more.
850 tsk
->flags
|= PF_EXITPIDONE
;
853 exit_io_context(tsk
);
855 if (tsk
->splice_pipe
)
856 free_pipe_info(tsk
->splice_pipe
);
858 if (tsk
->task_frag
.page
)
859 put_page(tsk
->task_frag
.page
);
861 validate_creds_for_do_exit(tsk
);
865 __this_cpu_add(dirty_throttle_leaks
, tsk
->nr_dirtied
);
869 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
870 * when the following two conditions become true.
871 * - There is race condition of mmap_sem (It is acquired by
873 * - SMI occurs before setting TASK_RUNINNG.
874 * (or hypervisor of virtual machine switches to other guest)
875 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
877 * To avoid it, we have to wait for releasing tsk->pi_lock which
878 * is held by try_to_wake_up()
881 raw_spin_unlock_wait(&tsk
->pi_lock
);
883 /* causes final put_task_struct in finish_task_switch(). */
884 tsk
->state
= TASK_DEAD
;
885 tsk
->flags
|= PF_NOFREEZE
; /* tell freezer to ignore us */
888 /* Avoid "noreturn function does return". */
890 cpu_relax(); /* For when BUG is null */
893 EXPORT_SYMBOL_GPL(do_exit
);
895 void complete_and_exit(struct completion
*comp
, long code
)
903 EXPORT_SYMBOL(complete_and_exit
);
905 SYSCALL_DEFINE1(exit
, int, error_code
)
907 do_exit((error_code
&0xff)<<8);
911 * Take down every thread in the group. This is called by fatal signals
912 * as well as by sys_exit_group (below).
915 do_group_exit(int exit_code
)
917 struct signal_struct
*sig
= current
->signal
;
919 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
921 if (signal_group_exit(sig
))
922 exit_code
= sig
->group_exit_code
;
923 else if (!thread_group_empty(current
)) {
924 struct sighand_struct
*const sighand
= current
->sighand
;
925 spin_lock_irq(&sighand
->siglock
);
926 if (signal_group_exit(sig
))
927 /* Another thread got here before we took the lock. */
928 exit_code
= sig
->group_exit_code
;
930 sig
->group_exit_code
= exit_code
;
931 sig
->flags
= SIGNAL_GROUP_EXIT
;
932 zap_other_threads(current
);
934 spin_unlock_irq(&sighand
->siglock
);
942 * this kills every thread in the thread group. Note that any externally
943 * wait4()-ing process will get the correct exit code - even if this
944 * thread is not the thread group leader.
946 SYSCALL_DEFINE1(exit_group
, int, error_code
)
948 do_group_exit((error_code
& 0xff) << 8);
954 enum pid_type wo_type
;
958 struct siginfo __user
*wo_info
;
960 struct rusage __user
*wo_rusage
;
962 wait_queue_t child_wait
;
967 struct pid
*task_pid_type(struct task_struct
*task
, enum pid_type type
)
969 if (type
!= PIDTYPE_PID
)
970 task
= task
->group_leader
;
971 return task
->pids
[type
].pid
;
974 static int eligible_pid(struct wait_opts
*wo
, struct task_struct
*p
)
976 return wo
->wo_type
== PIDTYPE_MAX
||
977 task_pid_type(p
, wo
->wo_type
) == wo
->wo_pid
;
980 static int eligible_child(struct wait_opts
*wo
, struct task_struct
*p
)
982 if (!eligible_pid(wo
, p
))
984 /* Wait for all children (clone and not) if __WALL is set;
985 * otherwise, wait for clone children *only* if __WCLONE is
986 * set; otherwise, wait for non-clone children *only*. (Note:
987 * A "clone" child here is one that reports to its parent
988 * using a signal other than SIGCHLD.) */
989 if (((p
->exit_signal
!= SIGCHLD
) ^ !!(wo
->wo_flags
& __WCLONE
))
990 && !(wo
->wo_flags
& __WALL
))
996 static int wait_noreap_copyout(struct wait_opts
*wo
, struct task_struct
*p
,
997 pid_t pid
, uid_t uid
, int why
, int status
)
999 struct siginfo __user
*infop
;
1000 int retval
= wo
->wo_rusage
1001 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1004 infop
= wo
->wo_info
;
1007 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1009 retval
= put_user(0, &infop
->si_errno
);
1011 retval
= put_user((short)why
, &infop
->si_code
);
1013 retval
= put_user(pid
, &infop
->si_pid
);
1015 retval
= put_user(uid
, &infop
->si_uid
);
1017 retval
= put_user(status
, &infop
->si_status
);
1025 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1026 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1027 * the lock and this task is uninteresting. If we return nonzero, we have
1028 * released the lock and the system call should return.
1030 static int wait_task_zombie(struct wait_opts
*wo
, struct task_struct
*p
)
1032 unsigned long state
;
1033 int retval
, status
, traced
;
1034 pid_t pid
= task_pid_vnr(p
);
1035 uid_t uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1036 struct siginfo __user
*infop
;
1038 if (!likely(wo
->wo_flags
& WEXITED
))
1041 if (unlikely(wo
->wo_flags
& WNOWAIT
)) {
1042 int exit_code
= p
->exit_code
;
1046 read_unlock(&tasklist_lock
);
1047 if ((exit_code
& 0x7f) == 0) {
1049 status
= exit_code
>> 8;
1051 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1052 status
= exit_code
& 0x7f;
1054 return wait_noreap_copyout(wo
, p
, pid
, uid
, why
, status
);
1058 * Try to move the task's state to DEAD
1059 * only one thread is allowed to do this:
1061 state
= xchg(&p
->exit_state
, EXIT_DEAD
);
1062 if (state
!= EXIT_ZOMBIE
) {
1063 BUG_ON(state
!= EXIT_DEAD
);
1067 traced
= ptrace_reparented(p
);
1069 * It can be ptraced but not reparented, check
1070 * thread_group_leader() to filter out sub-threads.
1072 if (likely(!traced
) && thread_group_leader(p
)) {
1073 struct signal_struct
*psig
;
1074 struct signal_struct
*sig
;
1075 unsigned long maxrss
;
1076 cputime_t tgutime
, tgstime
;
1079 * The resource counters for the group leader are in its
1080 * own task_struct. Those for dead threads in the group
1081 * are in its signal_struct, as are those for the child
1082 * processes it has previously reaped. All these
1083 * accumulate in the parent's signal_struct c* fields.
1085 * We don't bother to take a lock here to protect these
1086 * p->signal fields, because they are only touched by
1087 * __exit_signal, which runs with tasklist_lock
1088 * write-locked anyway, and so is excluded here. We do
1089 * need to protect the access to parent->signal fields,
1090 * as other threads in the parent group can be right
1091 * here reaping other children at the same time.
1093 * We use thread_group_cputime_adjusted() to get times for the thread
1094 * group, which consolidates times for all threads in the
1095 * group including the group leader.
1097 thread_group_cputime_adjusted(p
, &tgutime
, &tgstime
);
1098 spin_lock_irq(&p
->real_parent
->sighand
->siglock
);
1099 psig
= p
->real_parent
->signal
;
1101 psig
->cutime
+= tgutime
+ sig
->cutime
;
1102 psig
->cstime
+= tgstime
+ sig
->cstime
;
1103 psig
->cgtime
+= task_gtime(p
) + sig
->gtime
+ sig
->cgtime
;
1105 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1107 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1109 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1111 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1113 task_io_get_inblock(p
) +
1114 sig
->inblock
+ sig
->cinblock
;
1116 task_io_get_oublock(p
) +
1117 sig
->oublock
+ sig
->coublock
;
1118 maxrss
= max(sig
->maxrss
, sig
->cmaxrss
);
1119 if (psig
->cmaxrss
< maxrss
)
1120 psig
->cmaxrss
= maxrss
;
1121 task_io_accounting_add(&psig
->ioac
, &p
->ioac
);
1122 task_io_accounting_add(&psig
->ioac
, &sig
->ioac
);
1123 spin_unlock_irq(&p
->real_parent
->sighand
->siglock
);
1127 * Now we are sure this task is interesting, and no other
1128 * thread can reap it because we set its state to EXIT_DEAD.
1130 read_unlock(&tasklist_lock
);
1132 retval
= wo
->wo_rusage
1133 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1134 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1135 ? p
->signal
->group_exit_code
: p
->exit_code
;
1136 if (!retval
&& wo
->wo_stat
)
1137 retval
= put_user(status
, wo
->wo_stat
);
1139 infop
= wo
->wo_info
;
1140 if (!retval
&& infop
)
1141 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1142 if (!retval
&& infop
)
1143 retval
= put_user(0, &infop
->si_errno
);
1144 if (!retval
&& infop
) {
1147 if ((status
& 0x7f) == 0) {
1151 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1154 retval
= put_user((short)why
, &infop
->si_code
);
1156 retval
= put_user(status
, &infop
->si_status
);
1158 if (!retval
&& infop
)
1159 retval
= put_user(pid
, &infop
->si_pid
);
1160 if (!retval
&& infop
)
1161 retval
= put_user(uid
, &infop
->si_uid
);
1166 write_lock_irq(&tasklist_lock
);
1167 /* We dropped tasklist, ptracer could die and untrace */
1170 * If this is not a sub-thread, notify the parent.
1171 * If parent wants a zombie, don't release it now.
1173 if (thread_group_leader(p
) &&
1174 !do_notify_parent(p
, p
->exit_signal
)) {
1175 p
->exit_state
= EXIT_ZOMBIE
;
1178 write_unlock_irq(&tasklist_lock
);
1186 static int *task_stopped_code(struct task_struct
*p
, bool ptrace
)
1189 if (task_is_stopped_or_traced(p
) &&
1190 !(p
->jobctl
& JOBCTL_LISTENING
))
1191 return &p
->exit_code
;
1193 if (p
->signal
->flags
& SIGNAL_STOP_STOPPED
)
1194 return &p
->signal
->group_exit_code
;
1200 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1202 * @ptrace: is the wait for ptrace
1203 * @p: task to wait for
1205 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1208 * read_lock(&tasklist_lock), which is released if return value is
1209 * non-zero. Also, grabs and releases @p->sighand->siglock.
1212 * 0 if wait condition didn't exist and search for other wait conditions
1213 * should continue. Non-zero return, -errno on failure and @p's pid on
1214 * success, implies that tasklist_lock is released and wait condition
1215 * search should terminate.
1217 static int wait_task_stopped(struct wait_opts
*wo
,
1218 int ptrace
, struct task_struct
*p
)
1220 struct siginfo __user
*infop
;
1221 int retval
, exit_code
, *p_code
, why
;
1222 uid_t uid
= 0; /* unneeded, required by compiler */
1226 * Traditionally we see ptrace'd stopped tasks regardless of options.
1228 if (!ptrace
&& !(wo
->wo_flags
& WUNTRACED
))
1231 if (!task_stopped_code(p
, ptrace
))
1235 spin_lock_irq(&p
->sighand
->siglock
);
1237 p_code
= task_stopped_code(p
, ptrace
);
1238 if (unlikely(!p_code
))
1241 exit_code
= *p_code
;
1245 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1248 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1250 spin_unlock_irq(&p
->sighand
->siglock
);
1255 * Now we are pretty sure this task is interesting.
1256 * Make sure it doesn't get reaped out from under us while we
1257 * give up the lock and then examine it below. We don't want to
1258 * keep holding onto the tasklist_lock while we call getrusage and
1259 * possibly take page faults for user memory.
1262 pid
= task_pid_vnr(p
);
1263 why
= ptrace
? CLD_TRAPPED
: CLD_STOPPED
;
1264 read_unlock(&tasklist_lock
);
1266 if (unlikely(wo
->wo_flags
& WNOWAIT
))
1267 return wait_noreap_copyout(wo
, p
, pid
, uid
, why
, exit_code
);
1269 retval
= wo
->wo_rusage
1270 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1271 if (!retval
&& wo
->wo_stat
)
1272 retval
= put_user((exit_code
<< 8) | 0x7f, wo
->wo_stat
);
1274 infop
= wo
->wo_info
;
1275 if (!retval
&& infop
)
1276 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1277 if (!retval
&& infop
)
1278 retval
= put_user(0, &infop
->si_errno
);
1279 if (!retval
&& infop
)
1280 retval
= put_user((short)why
, &infop
->si_code
);
1281 if (!retval
&& infop
)
1282 retval
= put_user(exit_code
, &infop
->si_status
);
1283 if (!retval
&& infop
)
1284 retval
= put_user(pid
, &infop
->si_pid
);
1285 if (!retval
&& infop
)
1286 retval
= put_user(uid
, &infop
->si_uid
);
1296 * Handle do_wait work for one task in a live, non-stopped state.
1297 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1298 * the lock and this task is uninteresting. If we return nonzero, we have
1299 * released the lock and the system call should return.
1301 static int wait_task_continued(struct wait_opts
*wo
, struct task_struct
*p
)
1307 if (!unlikely(wo
->wo_flags
& WCONTINUED
))
1310 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1313 spin_lock_irq(&p
->sighand
->siglock
);
1314 /* Re-check with the lock held. */
1315 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1316 spin_unlock_irq(&p
->sighand
->siglock
);
1319 if (!unlikely(wo
->wo_flags
& WNOWAIT
))
1320 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1321 uid
= from_kuid_munged(current_user_ns(), task_uid(p
));
1322 spin_unlock_irq(&p
->sighand
->siglock
);
1324 pid
= task_pid_vnr(p
);
1326 read_unlock(&tasklist_lock
);
1329 retval
= wo
->wo_rusage
1330 ? getrusage(p
, RUSAGE_BOTH
, wo
->wo_rusage
) : 0;
1332 if (!retval
&& wo
->wo_stat
)
1333 retval
= put_user(0xffff, wo
->wo_stat
);
1337 retval
= wait_noreap_copyout(wo
, p
, pid
, uid
,
1338 CLD_CONTINUED
, SIGCONT
);
1339 BUG_ON(retval
== 0);
1346 * Consider @p for a wait by @parent.
1348 * -ECHILD should be in ->notask_error before the first call.
1349 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1350 * Returns zero if the search for a child should continue;
1351 * then ->notask_error is 0 if @p is an eligible child,
1352 * or another error from security_task_wait(), or still -ECHILD.
1354 static int wait_consider_task(struct wait_opts
*wo
, int ptrace
,
1355 struct task_struct
*p
)
1357 int ret
= eligible_child(wo
, p
);
1361 ret
= security_task_wait(p
);
1362 if (unlikely(ret
< 0)) {
1364 * If we have not yet seen any eligible child,
1365 * then let this error code replace -ECHILD.
1366 * A permission error will give the user a clue
1367 * to look for security policy problems, rather
1368 * than for mysterious wait bugs.
1370 if (wo
->notask_error
)
1371 wo
->notask_error
= ret
;
1375 /* dead body doesn't have much to contribute */
1376 if (unlikely(p
->exit_state
== EXIT_DEAD
)) {
1378 * But do not ignore this task until the tracer does
1379 * wait_task_zombie()->do_notify_parent().
1381 if (likely(!ptrace
) && unlikely(ptrace_reparented(p
)))
1382 wo
->notask_error
= 0;
1387 if (p
->exit_state
== EXIT_ZOMBIE
) {
1389 * A zombie ptracee is only visible to its ptracer.
1390 * Notification and reaping will be cascaded to the real
1391 * parent when the ptracer detaches.
1393 if (likely(!ptrace
) && unlikely(p
->ptrace
)) {
1394 /* it will become visible, clear notask_error */
1395 wo
->notask_error
= 0;
1399 /* we don't reap group leaders with subthreads */
1400 if (!delay_group_leader(p
))
1401 return wait_task_zombie(wo
, p
);
1404 * Allow access to stopped/continued state via zombie by
1405 * falling through. Clearing of notask_error is complex.
1409 * If WEXITED is set, notask_error should naturally be
1410 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1411 * so, if there are live subthreads, there are events to
1412 * wait for. If all subthreads are dead, it's still safe
1413 * to clear - this function will be called again in finite
1414 * amount time once all the subthreads are released and
1415 * will then return without clearing.
1419 * Stopped state is per-task and thus can't change once the
1420 * target task dies. Only continued and exited can happen.
1421 * Clear notask_error if WCONTINUED | WEXITED.
1423 if (likely(!ptrace
) || (wo
->wo_flags
& (WCONTINUED
| WEXITED
)))
1424 wo
->notask_error
= 0;
1427 * If @p is ptraced by a task in its real parent's group,
1428 * hide group stop/continued state when looking at @p as
1429 * the real parent; otherwise, a single stop can be
1430 * reported twice as group and ptrace stops.
1432 * If a ptracer wants to distinguish the two events for its
1433 * own children, it should create a separate process which
1434 * takes the role of real parent.
1436 if (likely(!ptrace
) && p
->ptrace
&& !ptrace_reparented(p
))
1440 * @p is alive and it's gonna stop, continue or exit, so
1441 * there always is something to wait for.
1443 wo
->notask_error
= 0;
1447 * Wait for stopped. Depending on @ptrace, different stopped state
1448 * is used and the two don't interact with each other.
1450 ret
= wait_task_stopped(wo
, ptrace
, p
);
1455 * Wait for continued. There's only one continued state and the
1456 * ptracer can consume it which can confuse the real parent. Don't
1457 * use WCONTINUED from ptracer. You don't need or want it.
1459 return wait_task_continued(wo
, p
);
1463 * Do the work of do_wait() for one thread in the group, @tsk.
1465 * -ECHILD should be in ->notask_error before the first call.
1466 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1467 * Returns zero if the search for a child should continue; then
1468 * ->notask_error is 0 if there were any eligible children,
1469 * or another error from security_task_wait(), or still -ECHILD.
1471 static int do_wait_thread(struct wait_opts
*wo
, struct task_struct
*tsk
)
1473 struct task_struct
*p
;
1475 list_for_each_entry(p
, &tsk
->children
, sibling
) {
1476 int ret
= wait_consider_task(wo
, 0, p
);
1484 static int ptrace_do_wait(struct wait_opts
*wo
, struct task_struct
*tsk
)
1486 struct task_struct
*p
;
1488 list_for_each_entry(p
, &tsk
->ptraced
, ptrace_entry
) {
1489 int ret
= wait_consider_task(wo
, 1, p
);
1497 static int child_wait_callback(wait_queue_t
*wait
, unsigned mode
,
1498 int sync
, void *key
)
1500 struct wait_opts
*wo
= container_of(wait
, struct wait_opts
,
1502 struct task_struct
*p
= key
;
1504 if (!eligible_pid(wo
, p
))
1507 if ((wo
->wo_flags
& __WNOTHREAD
) && wait
->private != p
->parent
)
1510 return default_wake_function(wait
, mode
, sync
, key
);
1513 void __wake_up_parent(struct task_struct
*p
, struct task_struct
*parent
)
1515 __wake_up_sync_key(&parent
->signal
->wait_chldexit
,
1516 TASK_INTERRUPTIBLE
, 1, p
);
1519 static long do_wait(struct wait_opts
*wo
)
1521 struct task_struct
*tsk
;
1524 trace_sched_process_wait(wo
->wo_pid
);
1526 init_waitqueue_func_entry(&wo
->child_wait
, child_wait_callback
);
1527 wo
->child_wait
.private = current
;
1528 add_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1531 * If there is nothing that can match our critiera just get out.
1532 * We will clear ->notask_error to zero if we see any child that
1533 * might later match our criteria, even if we are not able to reap
1536 wo
->notask_error
= -ECHILD
;
1537 if ((wo
->wo_type
< PIDTYPE_MAX
) &&
1538 (!wo
->wo_pid
|| hlist_empty(&wo
->wo_pid
->tasks
[wo
->wo_type
])))
1541 set_current_state(TASK_INTERRUPTIBLE
);
1542 read_lock(&tasklist_lock
);
1545 retval
= do_wait_thread(wo
, tsk
);
1549 retval
= ptrace_do_wait(wo
, tsk
);
1553 if (wo
->wo_flags
& __WNOTHREAD
)
1555 } while_each_thread(current
, tsk
);
1556 read_unlock(&tasklist_lock
);
1559 retval
= wo
->notask_error
;
1560 if (!retval
&& !(wo
->wo_flags
& WNOHANG
)) {
1561 retval
= -ERESTARTSYS
;
1562 if (!signal_pending(current
)) {
1568 __set_current_state(TASK_RUNNING
);
1569 remove_wait_queue(¤t
->signal
->wait_chldexit
, &wo
->child_wait
);
1573 SYSCALL_DEFINE5(waitid
, int, which
, pid_t
, upid
, struct siginfo __user
*,
1574 infop
, int, options
, struct rusage __user
*, ru
)
1576 struct wait_opts wo
;
1577 struct pid
*pid
= NULL
;
1581 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
))
1583 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1596 type
= PIDTYPE_PGID
;
1604 if (type
< PIDTYPE_MAX
)
1605 pid
= find_get_pid(upid
);
1609 wo
.wo_flags
= options
;
1619 * For a WNOHANG return, clear out all the fields
1620 * we would set so the user can easily tell the
1624 ret
= put_user(0, &infop
->si_signo
);
1626 ret
= put_user(0, &infop
->si_errno
);
1628 ret
= put_user(0, &infop
->si_code
);
1630 ret
= put_user(0, &infop
->si_pid
);
1632 ret
= put_user(0, &infop
->si_uid
);
1634 ret
= put_user(0, &infop
->si_status
);
1641 SYSCALL_DEFINE4(wait4
, pid_t
, upid
, int __user
*, stat_addr
,
1642 int, options
, struct rusage __user
*, ru
)
1644 struct wait_opts wo
;
1645 struct pid
*pid
= NULL
;
1649 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1650 __WNOTHREAD
|__WCLONE
|__WALL
))
1655 else if (upid
< 0) {
1656 type
= PIDTYPE_PGID
;
1657 pid
= find_get_pid(-upid
);
1658 } else if (upid
== 0) {
1659 type
= PIDTYPE_PGID
;
1660 pid
= get_task_pid(current
, PIDTYPE_PGID
);
1661 } else /* upid > 0 */ {
1663 pid
= find_get_pid(upid
);
1668 wo
.wo_flags
= options
| WEXITED
;
1670 wo
.wo_stat
= stat_addr
;
1678 #ifdef __ARCH_WANT_SYS_WAITPID
1681 * sys_waitpid() remains for compatibility. waitpid() should be
1682 * implemented by calling sys_wait4() from libc.a.
1684 SYSCALL_DEFINE3(waitpid
, pid_t
, pid
, int __user
*, stat_addr
, int, options
)
1686 return sys_wait4(pid
, stat_addr
, options
, NULL
);