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/mnt_namespace.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/binfmts.h>
23 #include <linux/nsproxy.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/ptrace.h>
26 #include <linux/profile.h>
27 #include <linux/mount.h>
28 #include <linux/proc_fs.h>
29 #include <linux/kthread.h>
30 #include <linux/mempolicy.h>
31 #include <linux/taskstats_kern.h>
32 #include <linux/delayacct.h>
33 #include <linux/cpuset.h>
34 #include <linux/syscalls.h>
35 #include <linux/signal.h>
36 #include <linux/posix-timers.h>
37 #include <linux/cn_proc.h>
38 #include <linux/mutex.h>
39 #include <linux/futex.h>
40 #include <linux/compat.h>
41 #include <linux/pipe_fs_i.h>
42 #include <linux/audit.h> /* for audit_free() */
43 #include <linux/resource.h>
44 #include <linux/blkdev.h>
45 #include <linux/task_io_accounting_ops.h>
47 #include <asm/uaccess.h>
48 #include <asm/unistd.h>
49 #include <asm/pgtable.h>
50 #include <asm/mmu_context.h>
52 extern void sem_exit (void);
54 static void exit_mm(struct task_struct
* tsk
);
56 static void __unhash_process(struct task_struct
*p
)
59 detach_pid(p
, PIDTYPE_PID
);
60 if (thread_group_leader(p
)) {
61 detach_pid(p
, PIDTYPE_PGID
);
62 detach_pid(p
, PIDTYPE_SID
);
64 list_del_rcu(&p
->tasks
);
65 __get_cpu_var(process_counts
)--;
67 list_del_rcu(&p
->thread_group
);
72 * This function expects the tasklist_lock write-locked.
74 static void __exit_signal(struct task_struct
*tsk
)
76 struct signal_struct
*sig
= tsk
->signal
;
77 struct sighand_struct
*sighand
;
80 BUG_ON(!atomic_read(&sig
->count
));
83 sighand
= rcu_dereference(tsk
->sighand
);
84 spin_lock(&sighand
->siglock
);
86 posix_cpu_timers_exit(tsk
);
87 if (atomic_dec_and_test(&sig
->count
))
88 posix_cpu_timers_exit_group(tsk
);
91 * If there is any task waiting for the group exit
94 if (sig
->group_exit_task
&& atomic_read(&sig
->count
) == sig
->notify_count
) {
95 wake_up_process(sig
->group_exit_task
);
96 sig
->group_exit_task
= NULL
;
98 if (tsk
== sig
->curr_target
)
99 sig
->curr_target
= next_thread(tsk
);
101 * Accumulate here the counters for all threads but the
102 * group leader as they die, so they can be added into
103 * the process-wide totals when those are taken.
104 * The group leader stays around as a zombie as long
105 * as there are other threads. When it gets reaped,
106 * the exit.c code will add its counts into these totals.
107 * We won't ever get here for the group leader, since it
108 * will have been the last reference on the signal_struct.
110 sig
->utime
= cputime_add(sig
->utime
, tsk
->utime
);
111 sig
->stime
= cputime_add(sig
->stime
, tsk
->stime
);
112 sig
->min_flt
+= tsk
->min_flt
;
113 sig
->maj_flt
+= tsk
->maj_flt
;
114 sig
->nvcsw
+= tsk
->nvcsw
;
115 sig
->nivcsw
+= tsk
->nivcsw
;
116 sig
->sched_time
+= tsk
->sched_time
;
117 sig
->inblock
+= task_io_get_inblock(tsk
);
118 sig
->oublock
+= task_io_get_oublock(tsk
);
119 sig
= NULL
; /* Marker for below. */
122 __unhash_process(tsk
);
126 spin_unlock(&sighand
->siglock
);
129 __cleanup_sighand(sighand
);
130 clear_tsk_thread_flag(tsk
,TIF_SIGPENDING
);
131 flush_sigqueue(&tsk
->pending
);
133 flush_sigqueue(&sig
->shared_pending
);
134 taskstats_tgid_free(sig
);
135 __cleanup_signal(sig
);
139 static void delayed_put_task_struct(struct rcu_head
*rhp
)
141 put_task_struct(container_of(rhp
, struct task_struct
, rcu
));
144 void release_task(struct task_struct
* p
)
146 struct task_struct
*leader
;
149 atomic_dec(&p
->user
->processes
);
150 write_lock_irq(&tasklist_lock
);
152 BUG_ON(!list_empty(&p
->ptrace_list
) || !list_empty(&p
->ptrace_children
));
156 * If we are the last non-leader member of the thread
157 * group, and the leader is zombie, then notify the
158 * group leader's parent process. (if it wants notification.)
161 leader
= p
->group_leader
;
162 if (leader
!= p
&& thread_group_empty(leader
) && leader
->exit_state
== EXIT_ZOMBIE
) {
163 BUG_ON(leader
->exit_signal
== -1);
164 do_notify_parent(leader
, leader
->exit_signal
);
166 * If we were the last child thread and the leader has
167 * exited already, and the leader's parent ignores SIGCHLD,
168 * then we are the one who should release the leader.
170 * do_notify_parent() will have marked it self-reaping in
173 zap_leader
= (leader
->exit_signal
== -1);
177 write_unlock_irq(&tasklist_lock
);
180 call_rcu(&p
->rcu
, delayed_put_task_struct
);
183 if (unlikely(zap_leader
))
188 * This checks not only the pgrp, but falls back on the pid if no
189 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
192 * The caller must hold rcu lock or the tasklist lock.
194 struct pid
*session_of_pgrp(struct pid
*pgrp
)
196 struct task_struct
*p
;
197 struct pid
*sid
= NULL
;
199 p
= pid_task(pgrp
, PIDTYPE_PGID
);
201 p
= pid_task(pgrp
, PIDTYPE_PID
);
203 sid
= task_session(p
);
209 * Determine if a process group is "orphaned", according to the POSIX
210 * definition in 2.2.2.52. Orphaned process groups are not to be affected
211 * by terminal-generated stop signals. Newly orphaned process groups are
212 * to receive a SIGHUP and a SIGCONT.
214 * "I ask you, have you ever known what it is to be an orphan?"
216 static int will_become_orphaned_pgrp(struct pid
*pgrp
, struct task_struct
*ignored_task
)
218 struct task_struct
*p
;
221 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
222 if (p
== ignored_task
224 || is_init(p
->real_parent
))
226 if (task_pgrp(p
->real_parent
) != pgrp
&&
227 task_session(p
->real_parent
) == task_session(p
)) {
231 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
232 return ret
; /* (sighing) "Often!" */
235 int is_current_pgrp_orphaned(void)
239 read_lock(&tasklist_lock
);
240 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
241 read_unlock(&tasklist_lock
);
246 static int has_stopped_jobs(struct pid
*pgrp
)
249 struct task_struct
*p
;
251 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
252 if (p
->state
!= TASK_STOPPED
)
256 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
261 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
263 * If a kernel thread is launched as a result of a system call, or if
264 * it ever exits, it should generally reparent itself to kthreadd so it
265 * isn't in the way of other processes and is correctly cleaned up on exit.
267 * The various task state such as scheduling policy and priority may have
268 * been inherited from a user process, so we reset them to sane values here.
270 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
272 static void reparent_to_kthreadd(void)
274 write_lock_irq(&tasklist_lock
);
276 ptrace_unlink(current
);
277 /* Reparent to init */
278 remove_parent(current
);
279 current
->real_parent
= current
->parent
= kthreadd_task
;
282 /* Set the exit signal to SIGCHLD so we signal init on exit */
283 current
->exit_signal
= SIGCHLD
;
285 if (!has_rt_policy(current
) && (task_nice(current
) < 0))
286 set_user_nice(current
, 0);
290 security_task_reparent_to_init(current
);
291 memcpy(current
->signal
->rlim
, init_task
.signal
->rlim
,
292 sizeof(current
->signal
->rlim
));
293 atomic_inc(&(INIT_USER
->__count
));
294 write_unlock_irq(&tasklist_lock
);
295 switch_uid(INIT_USER
);
298 void __set_special_pids(pid_t session
, pid_t pgrp
)
300 struct task_struct
*curr
= current
->group_leader
;
302 if (process_session(curr
) != session
) {
303 detach_pid(curr
, PIDTYPE_SID
);
304 set_signal_session(curr
->signal
, session
);
305 attach_pid(curr
, PIDTYPE_SID
, session
);
307 if (process_group(curr
) != pgrp
) {
308 detach_pid(curr
, PIDTYPE_PGID
);
309 curr
->signal
->pgrp
= pgrp
;
310 attach_pid(curr
, PIDTYPE_PGID
, pgrp
);
314 static void set_special_pids(pid_t session
, pid_t pgrp
)
316 write_lock_irq(&tasklist_lock
);
317 __set_special_pids(session
, pgrp
);
318 write_unlock_irq(&tasklist_lock
);
322 * Let kernel threads use this to say that they
323 * allow a certain signal (since daemonize() will
324 * have disabled all of them by default).
326 int allow_signal(int sig
)
328 if (!valid_signal(sig
) || sig
< 1)
331 spin_lock_irq(¤t
->sighand
->siglock
);
332 sigdelset(¤t
->blocked
, sig
);
334 /* Kernel threads handle their own signals.
335 Let the signal code know it'll be handled, so
336 that they don't get converted to SIGKILL or
337 just silently dropped */
338 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= (void __user
*)2;
341 spin_unlock_irq(¤t
->sighand
->siglock
);
345 EXPORT_SYMBOL(allow_signal
);
347 int disallow_signal(int sig
)
349 if (!valid_signal(sig
) || sig
< 1)
352 spin_lock_irq(¤t
->sighand
->siglock
);
353 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= SIG_IGN
;
355 spin_unlock_irq(¤t
->sighand
->siglock
);
359 EXPORT_SYMBOL(disallow_signal
);
362 * Put all the gunge required to become a kernel thread without
363 * attached user resources in one place where it belongs.
366 void daemonize(const char *name
, ...)
369 struct fs_struct
*fs
;
372 va_start(args
, name
);
373 vsnprintf(current
->comm
, sizeof(current
->comm
), name
, args
);
377 * If we were started as result of loading a module, close all of the
378 * user space pages. We don't need them, and if we didn't close them
379 * they would be locked into memory.
383 set_special_pids(1, 1);
384 proc_clear_tty(current
);
386 /* Block and flush all signals */
387 sigfillset(&blocked
);
388 sigprocmask(SIG_BLOCK
, &blocked
, NULL
);
389 flush_signals(current
);
391 /* Become as one with the init task */
393 exit_fs(current
); /* current->fs->count--; */
396 atomic_inc(&fs
->count
);
398 exit_task_namespaces(current
);
399 current
->nsproxy
= init_task
.nsproxy
;
400 get_task_namespaces(current
);
403 current
->files
= init_task
.files
;
404 atomic_inc(¤t
->files
->count
);
406 reparent_to_kthreadd();
409 EXPORT_SYMBOL(daemonize
);
411 static void close_files(struct files_struct
* files
)
419 * It is safe to dereference the fd table without RCU or
420 * ->file_lock because this is the last reference to the
423 fdt
= files_fdtable(files
);
427 if (i
>= fdt
->max_fds
)
429 set
= fdt
->open_fds
->fds_bits
[j
++];
432 struct file
* file
= xchg(&fdt
->fd
[i
], NULL
);
434 filp_close(file
, files
);
444 struct files_struct
*get_files_struct(struct task_struct
*task
)
446 struct files_struct
*files
;
451 atomic_inc(&files
->count
);
457 void fastcall
put_files_struct(struct files_struct
*files
)
461 if (atomic_dec_and_test(&files
->count
)) {
464 * Free the fd and fdset arrays if we expanded them.
465 * If the fdtable was embedded, pass files for freeing
466 * at the end of the RCU grace period. Otherwise,
467 * you can free files immediately.
469 fdt
= files_fdtable(files
);
470 if (fdt
!= &files
->fdtab
)
471 kmem_cache_free(files_cachep
, files
);
476 EXPORT_SYMBOL(put_files_struct
);
478 void reset_files_struct(struct task_struct
*tsk
, struct files_struct
*files
)
480 struct files_struct
*old
;
486 put_files_struct(old
);
488 EXPORT_SYMBOL(reset_files_struct
);
490 static inline void __exit_files(struct task_struct
*tsk
)
492 struct files_struct
* files
= tsk
->files
;
498 put_files_struct(files
);
502 void exit_files(struct task_struct
*tsk
)
507 static inline void __put_fs_struct(struct fs_struct
*fs
)
509 /* No need to hold fs->lock if we are killing it */
510 if (atomic_dec_and_test(&fs
->count
)) {
517 mntput(fs
->altrootmnt
);
519 kmem_cache_free(fs_cachep
, fs
);
523 void put_fs_struct(struct fs_struct
*fs
)
528 static inline void __exit_fs(struct task_struct
*tsk
)
530 struct fs_struct
* fs
= tsk
->fs
;
540 void exit_fs(struct task_struct
*tsk
)
545 EXPORT_SYMBOL_GPL(exit_fs
);
548 * Turn us into a lazy TLB process if we
551 static void exit_mm(struct task_struct
* tsk
)
553 struct mm_struct
*mm
= tsk
->mm
;
559 * Serialize with any possible pending coredump.
560 * We must hold mmap_sem around checking core_waiters
561 * and clearing tsk->mm. The core-inducing thread
562 * will increment core_waiters for each thread in the
563 * group with ->mm != NULL.
565 down_read(&mm
->mmap_sem
);
566 if (mm
->core_waiters
) {
567 up_read(&mm
->mmap_sem
);
568 down_write(&mm
->mmap_sem
);
569 if (!--mm
->core_waiters
)
570 complete(mm
->core_startup_done
);
571 up_write(&mm
->mmap_sem
);
573 wait_for_completion(&mm
->core_done
);
574 down_read(&mm
->mmap_sem
);
576 atomic_inc(&mm
->mm_count
);
577 BUG_ON(mm
!= tsk
->active_mm
);
578 /* more a memory barrier than a real lock */
581 up_read(&mm
->mmap_sem
);
582 enter_lazy_tlb(mm
, current
);
588 choose_new_parent(struct task_struct
*p
, struct task_struct
*reaper
)
591 * Make sure we're not reparenting to ourselves and that
592 * the parent is not a zombie.
594 BUG_ON(p
== reaper
|| reaper
->exit_state
);
595 p
->real_parent
= reaper
;
599 reparent_thread(struct task_struct
*p
, struct task_struct
*father
, int traced
)
601 if (p
->pdeath_signal
)
602 /* We already hold the tasklist_lock here. */
603 group_send_sig_info(p
->pdeath_signal
, SEND_SIG_NOINFO
, p
);
605 /* Move the child from its dying parent to the new one. */
606 if (unlikely(traced
)) {
607 /* Preserve ptrace links if someone else is tracing this child. */
608 list_del_init(&p
->ptrace_list
);
609 if (p
->parent
!= p
->real_parent
)
610 list_add(&p
->ptrace_list
, &p
->real_parent
->ptrace_children
);
612 /* If this child is being traced, then we're the one tracing it
613 * anyway, so let go of it.
617 p
->parent
= p
->real_parent
;
620 if (p
->state
== TASK_TRACED
) {
622 * If it was at a trace stop, turn it into
623 * a normal stop since it's no longer being
630 /* If this is a threaded reparent there is no need to
631 * notify anyone anything has happened.
633 if (p
->real_parent
->group_leader
== father
->group_leader
)
636 /* We don't want people slaying init. */
637 if (p
->exit_signal
!= -1)
638 p
->exit_signal
= SIGCHLD
;
640 /* If we'd notified the old parent about this child's death,
641 * also notify the new parent.
643 if (!traced
&& p
->exit_state
== EXIT_ZOMBIE
&&
644 p
->exit_signal
!= -1 && thread_group_empty(p
))
645 do_notify_parent(p
, p
->exit_signal
);
648 * process group orphan check
649 * Case ii: Our child is in a different pgrp
650 * than we are, and it was the only connection
651 * outside, so the child pgrp is now orphaned.
653 if ((task_pgrp(p
) != task_pgrp(father
)) &&
654 (task_session(p
) == task_session(father
))) {
655 struct pid
*pgrp
= task_pgrp(p
);
657 if (will_become_orphaned_pgrp(pgrp
, NULL
) &&
658 has_stopped_jobs(pgrp
)) {
659 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
660 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
666 * When we die, we re-parent all our children.
667 * Try to give them to another thread in our thread
668 * group, and if no such member exists, give it to
669 * the child reaper process (ie "init") in our pid
673 forget_original_parent(struct task_struct
*father
, struct list_head
*to_release
)
675 struct task_struct
*p
, *reaper
= father
;
676 struct list_head
*_p
, *_n
;
679 reaper
= next_thread(reaper
);
680 if (reaper
== father
) {
681 reaper
= child_reaper(father
);
684 } while (reaper
->exit_state
);
687 * There are only two places where our children can be:
689 * - in our child list
690 * - in our ptraced child list
692 * Search them and reparent children.
694 list_for_each_safe(_p
, _n
, &father
->children
) {
696 p
= list_entry(_p
, struct task_struct
, sibling
);
700 /* if father isn't the real parent, then ptrace must be enabled */
701 BUG_ON(father
!= p
->real_parent
&& !ptrace
);
703 if (father
== p
->real_parent
) {
704 /* reparent with a reaper, real father it's us */
705 choose_new_parent(p
, reaper
);
706 reparent_thread(p
, father
, 0);
708 /* reparent ptraced task to its real parent */
710 if (p
->exit_state
== EXIT_ZOMBIE
&& p
->exit_signal
!= -1 &&
711 thread_group_empty(p
))
712 do_notify_parent(p
, p
->exit_signal
);
716 * if the ptraced child is a zombie with exit_signal == -1
717 * we must collect it before we exit, or it will remain
718 * zombie forever since we prevented it from self-reap itself
719 * while it was being traced by us, to be able to see it in wait4.
721 if (unlikely(ptrace
&& p
->exit_state
== EXIT_ZOMBIE
&& p
->exit_signal
== -1))
722 list_add(&p
->ptrace_list
, to_release
);
724 list_for_each_safe(_p
, _n
, &father
->ptrace_children
) {
725 p
= list_entry(_p
, struct task_struct
, ptrace_list
);
726 choose_new_parent(p
, reaper
);
727 reparent_thread(p
, father
, 1);
732 * Send signals to all our closest relatives so that they know
733 * to properly mourn us..
735 static void exit_notify(struct task_struct
*tsk
)
738 struct task_struct
*t
;
739 struct list_head ptrace_dead
, *_p
, *_n
;
742 if (signal_pending(tsk
) && !(tsk
->signal
->flags
& SIGNAL_GROUP_EXIT
)
743 && !thread_group_empty(tsk
)) {
745 * This occurs when there was a race between our exit
746 * syscall and a group signal choosing us as the one to
747 * wake up. It could be that we are the only thread
748 * alerted to check for pending signals, but another thread
749 * should be woken now to take the signal since we will not.
750 * Now we'll wake all the threads in the group just to make
751 * sure someone gets all the pending signals.
753 read_lock(&tasklist_lock
);
754 spin_lock_irq(&tsk
->sighand
->siglock
);
755 for (t
= next_thread(tsk
); t
!= tsk
; t
= next_thread(t
))
756 if (!signal_pending(t
) && !(t
->flags
& PF_EXITING
)) {
757 recalc_sigpending_tsk(t
);
758 if (signal_pending(t
))
759 signal_wake_up(t
, 0);
761 spin_unlock_irq(&tsk
->sighand
->siglock
);
762 read_unlock(&tasklist_lock
);
765 write_lock_irq(&tasklist_lock
);
768 * This does two things:
770 * A. Make init inherit all the child processes
771 * B. Check to see if any process groups have become orphaned
772 * as a result of our exiting, and if they have any stopped
773 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
776 INIT_LIST_HEAD(&ptrace_dead
);
777 forget_original_parent(tsk
, &ptrace_dead
);
778 BUG_ON(!list_empty(&tsk
->children
));
779 BUG_ON(!list_empty(&tsk
->ptrace_children
));
782 * Check to see if any process groups have become orphaned
783 * as a result of our exiting, and if they have any stopped
784 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
786 * Case i: Our father is in a different pgrp than we are
787 * and we were the only connection outside, so our pgrp
788 * is about to become orphaned.
791 t
= tsk
->real_parent
;
793 pgrp
= task_pgrp(tsk
);
794 if ((task_pgrp(t
) != pgrp
) &&
795 (task_session(t
) == task_session(tsk
)) &&
796 will_become_orphaned_pgrp(pgrp
, tsk
) &&
797 has_stopped_jobs(pgrp
)) {
798 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
799 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
802 /* Let father know we died
804 * Thread signals are configurable, but you aren't going to use
805 * that to send signals to arbitary processes.
806 * That stops right now.
808 * If the parent exec id doesn't match the exec id we saved
809 * when we started then we know the parent has changed security
812 * If our self_exec id doesn't match our parent_exec_id then
813 * we have changed execution domain as these two values started
814 * the same after a fork.
818 if (tsk
->exit_signal
!= SIGCHLD
&& tsk
->exit_signal
!= -1 &&
819 ( tsk
->parent_exec_id
!= t
->self_exec_id
||
820 tsk
->self_exec_id
!= tsk
->parent_exec_id
)
821 && !capable(CAP_KILL
))
822 tsk
->exit_signal
= SIGCHLD
;
825 /* If something other than our normal parent is ptracing us, then
826 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
827 * only has special meaning to our real parent.
829 if (tsk
->exit_signal
!= -1 && thread_group_empty(tsk
)) {
830 int signal
= tsk
->parent
== tsk
->real_parent
? tsk
->exit_signal
: SIGCHLD
;
831 do_notify_parent(tsk
, signal
);
832 } else if (tsk
->ptrace
) {
833 do_notify_parent(tsk
, SIGCHLD
);
837 if (tsk
->exit_signal
== -1 &&
838 (likely(tsk
->ptrace
== 0) ||
839 unlikely(tsk
->parent
->signal
->flags
& SIGNAL_GROUP_EXIT
)))
841 tsk
->exit_state
= state
;
843 write_unlock_irq(&tasklist_lock
);
845 list_for_each_safe(_p
, _n
, &ptrace_dead
) {
847 t
= list_entry(_p
, struct task_struct
, ptrace_list
);
851 /* If the process is dead, release it - nobody will wait for it */
852 if (state
== EXIT_DEAD
)
856 fastcall NORET_TYPE
void do_exit(long code
)
858 struct task_struct
*tsk
= current
;
861 profile_task_exit(tsk
);
863 WARN_ON(atomic_read(&tsk
->fs_excl
));
865 if (unlikely(in_interrupt()))
866 panic("Aiee, killing interrupt handler!");
867 if (unlikely(!tsk
->pid
))
868 panic("Attempted to kill the idle task!");
869 if (unlikely(tsk
== child_reaper(tsk
))) {
870 if (tsk
->nsproxy
->pid_ns
!= &init_pid_ns
)
871 tsk
->nsproxy
->pid_ns
->child_reaper
= init_pid_ns
.child_reaper
;
873 panic("Attempted to kill init!");
877 if (unlikely(current
->ptrace
& PT_TRACE_EXIT
)) {
878 current
->ptrace_message
= code
;
879 ptrace_notify((PTRACE_EVENT_EXIT
<< 8) | SIGTRAP
);
883 * We're taking recursive faults here in do_exit. Safest is to just
884 * leave this task alone and wait for reboot.
886 if (unlikely(tsk
->flags
& PF_EXITING
)) {
888 "Fixing recursive fault but reboot is needed!\n");
891 set_current_state(TASK_UNINTERRUPTIBLE
);
895 tsk
->flags
|= PF_EXITING
;
897 if (unlikely(in_atomic()))
898 printk(KERN_INFO
"note: %s[%d] exited with preempt_count %d\n",
899 current
->comm
, current
->pid
,
902 acct_update_integrals(tsk
);
904 update_hiwater_rss(tsk
->mm
);
905 update_hiwater_vm(tsk
->mm
);
907 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
909 hrtimer_cancel(&tsk
->signal
->real_timer
);
910 exit_itimers(tsk
->signal
);
912 acct_collect(code
, group_dead
);
913 if (unlikely(tsk
->robust_list
))
914 exit_robust_list(tsk
);
915 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
916 if (unlikely(tsk
->compat_robust_list
))
917 compat_exit_robust_list(tsk
);
919 if (unlikely(tsk
->audit_context
))
922 taskstats_exit(tsk
, group_dead
);
935 if (group_dead
&& tsk
->signal
->leader
)
936 disassociate_ctty(1);
938 module_put(task_thread_info(tsk
)->exec_domain
->module
);
940 module_put(tsk
->binfmt
->module
);
942 tsk
->exit_code
= code
;
943 proc_exit_connector(tsk
);
944 exit_task_namespaces(tsk
);
947 mpol_free(tsk
->mempolicy
);
948 tsk
->mempolicy
= NULL
;
951 * This must happen late, after the PID is not
954 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
955 exit_pi_state_list(tsk
);
956 if (unlikely(current
->pi_state_cache
))
957 kfree(current
->pi_state_cache
);
959 * Make sure we are holding no locks:
961 debug_check_no_locks_held(tsk
);
966 if (tsk
->splice_pipe
)
967 __free_pipe_info(tsk
->splice_pipe
);
970 /* causes final put_task_struct in finish_task_switch(). */
971 tsk
->state
= TASK_DEAD
;
975 /* Avoid "noreturn function does return". */
977 cpu_relax(); /* For when BUG is null */
980 EXPORT_SYMBOL_GPL(do_exit
);
982 NORET_TYPE
void complete_and_exit(struct completion
*comp
, long code
)
990 EXPORT_SYMBOL(complete_and_exit
);
992 asmlinkage
long sys_exit(int error_code
)
994 do_exit((error_code
&0xff)<<8);
998 * Take down every thread in the group. This is called by fatal signals
999 * as well as by sys_exit_group (below).
1002 do_group_exit(int exit_code
)
1004 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
1006 if (current
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1007 exit_code
= current
->signal
->group_exit_code
;
1008 else if (!thread_group_empty(current
)) {
1009 struct signal_struct
*const sig
= current
->signal
;
1010 struct sighand_struct
*const sighand
= current
->sighand
;
1011 spin_lock_irq(&sighand
->siglock
);
1012 if (sig
->flags
& SIGNAL_GROUP_EXIT
)
1013 /* Another thread got here before we took the lock. */
1014 exit_code
= sig
->group_exit_code
;
1016 sig
->group_exit_code
= exit_code
;
1017 zap_other_threads(current
);
1019 spin_unlock_irq(&sighand
->siglock
);
1027 * this kills every thread in the thread group. Note that any externally
1028 * wait4()-ing process will get the correct exit code - even if this
1029 * thread is not the thread group leader.
1031 asmlinkage
void sys_exit_group(int error_code
)
1033 do_group_exit((error_code
& 0xff) << 8);
1036 static int eligible_child(pid_t pid
, int options
, struct task_struct
*p
)
1044 if (process_group(p
) != process_group(current
))
1046 } else if (pid
!= -1) {
1047 if (process_group(p
) != -pid
)
1052 * Do not consider detached threads that are
1055 if (p
->exit_signal
== -1 && !p
->ptrace
)
1058 /* Wait for all children (clone and not) if __WALL is set;
1059 * otherwise, wait for clone children *only* if __WCLONE is
1060 * set; otherwise, wait for non-clone children *only*. (Note:
1061 * A "clone" child here is one that reports to its parent
1062 * using a signal other than SIGCHLD.) */
1063 if (((p
->exit_signal
!= SIGCHLD
) ^ ((options
& __WCLONE
) != 0))
1064 && !(options
& __WALL
))
1067 * Do not consider thread group leaders that are
1068 * in a non-empty thread group:
1070 if (delay_group_leader(p
))
1073 err
= security_task_wait(p
);
1080 static int wait_noreap_copyout(struct task_struct
*p
, pid_t pid
, uid_t uid
,
1081 int why
, int status
,
1082 struct siginfo __user
*infop
,
1083 struct rusage __user
*rusagep
)
1085 int retval
= rusagep
? getrusage(p
, RUSAGE_BOTH
, rusagep
) : 0;
1089 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1091 retval
= put_user(0, &infop
->si_errno
);
1093 retval
= put_user((short)why
, &infop
->si_code
);
1095 retval
= put_user(pid
, &infop
->si_pid
);
1097 retval
= put_user(uid
, &infop
->si_uid
);
1099 retval
= put_user(status
, &infop
->si_status
);
1106 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1107 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1108 * the lock and this task is uninteresting. If we return nonzero, we have
1109 * released the lock and the system call should return.
1111 static int wait_task_zombie(struct task_struct
*p
, int noreap
,
1112 struct siginfo __user
*infop
,
1113 int __user
*stat_addr
, struct rusage __user
*ru
)
1115 unsigned long state
;
1119 if (unlikely(noreap
)) {
1122 int exit_code
= p
->exit_code
;
1125 if (unlikely(p
->exit_state
!= EXIT_ZOMBIE
))
1127 if (unlikely(p
->exit_signal
== -1 && p
->ptrace
== 0))
1130 read_unlock(&tasklist_lock
);
1131 if ((exit_code
& 0x7f) == 0) {
1133 status
= exit_code
>> 8;
1135 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1136 status
= exit_code
& 0x7f;
1138 return wait_noreap_copyout(p
, pid
, uid
, why
,
1143 * Try to move the task's state to DEAD
1144 * only one thread is allowed to do this:
1146 state
= xchg(&p
->exit_state
, EXIT_DEAD
);
1147 if (state
!= EXIT_ZOMBIE
) {
1148 BUG_ON(state
!= EXIT_DEAD
);
1151 if (unlikely(p
->exit_signal
== -1 && p
->ptrace
== 0)) {
1153 * This can only happen in a race with a ptraced thread
1154 * dying on another processor.
1159 if (likely(p
->real_parent
== p
->parent
) && likely(p
->signal
)) {
1160 struct signal_struct
*psig
;
1161 struct signal_struct
*sig
;
1164 * The resource counters for the group leader are in its
1165 * own task_struct. Those for dead threads in the group
1166 * are in its signal_struct, as are those for the child
1167 * processes it has previously reaped. All these
1168 * accumulate in the parent's signal_struct c* fields.
1170 * We don't bother to take a lock here to protect these
1171 * p->signal fields, because they are only touched by
1172 * __exit_signal, which runs with tasklist_lock
1173 * write-locked anyway, and so is excluded here. We do
1174 * need to protect the access to p->parent->signal fields,
1175 * as other threads in the parent group can be right
1176 * here reaping other children at the same time.
1178 spin_lock_irq(&p
->parent
->sighand
->siglock
);
1179 psig
= p
->parent
->signal
;
1182 cputime_add(psig
->cutime
,
1183 cputime_add(p
->utime
,
1184 cputime_add(sig
->utime
,
1187 cputime_add(psig
->cstime
,
1188 cputime_add(p
->stime
,
1189 cputime_add(sig
->stime
,
1192 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1194 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1196 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1198 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1200 task_io_get_inblock(p
) +
1201 sig
->inblock
+ sig
->cinblock
;
1203 task_io_get_oublock(p
) +
1204 sig
->oublock
+ sig
->coublock
;
1205 spin_unlock_irq(&p
->parent
->sighand
->siglock
);
1209 * Now we are sure this task is interesting, and no other
1210 * thread can reap it because we set its state to EXIT_DEAD.
1212 read_unlock(&tasklist_lock
);
1214 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1215 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1216 ? p
->signal
->group_exit_code
: p
->exit_code
;
1217 if (!retval
&& stat_addr
)
1218 retval
= put_user(status
, stat_addr
);
1219 if (!retval
&& infop
)
1220 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1221 if (!retval
&& infop
)
1222 retval
= put_user(0, &infop
->si_errno
);
1223 if (!retval
&& infop
) {
1226 if ((status
& 0x7f) == 0) {
1230 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1233 retval
= put_user((short)why
, &infop
->si_code
);
1235 retval
= put_user(status
, &infop
->si_status
);
1237 if (!retval
&& infop
)
1238 retval
= put_user(p
->pid
, &infop
->si_pid
);
1239 if (!retval
&& infop
)
1240 retval
= put_user(p
->uid
, &infop
->si_uid
);
1242 // TODO: is this safe?
1243 p
->exit_state
= EXIT_ZOMBIE
;
1247 if (p
->real_parent
!= p
->parent
) {
1248 write_lock_irq(&tasklist_lock
);
1249 /* Double-check with lock held. */
1250 if (p
->real_parent
!= p
->parent
) {
1252 // TODO: is this safe?
1253 p
->exit_state
= EXIT_ZOMBIE
;
1255 * If this is not a detached task, notify the parent.
1256 * If it's still not detached after that, don't release
1259 if (p
->exit_signal
!= -1) {
1260 do_notify_parent(p
, p
->exit_signal
);
1261 if (p
->exit_signal
!= -1)
1265 write_unlock_irq(&tasklist_lock
);
1274 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1275 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1276 * the lock and this task is uninteresting. If we return nonzero, we have
1277 * released the lock and the system call should return.
1279 static int wait_task_stopped(struct task_struct
*p
, int delayed_group_leader
,
1280 int noreap
, struct siginfo __user
*infop
,
1281 int __user
*stat_addr
, struct rusage __user
*ru
)
1283 int retval
, exit_code
;
1287 if (delayed_group_leader
&& !(p
->ptrace
& PT_PTRACED
) &&
1288 p
->signal
&& p
->signal
->group_stop_count
> 0)
1290 * A group stop is in progress and this is the group leader.
1291 * We won't report until all threads have stopped.
1296 * Now we are pretty sure this task is interesting.
1297 * Make sure it doesn't get reaped out from under us while we
1298 * give up the lock and then examine it below. We don't want to
1299 * keep holding onto the tasklist_lock while we call getrusage and
1300 * possibly take page faults for user memory.
1303 read_unlock(&tasklist_lock
);
1305 if (unlikely(noreap
)) {
1308 int why
= (p
->ptrace
& PT_PTRACED
) ? CLD_TRAPPED
: CLD_STOPPED
;
1310 exit_code
= p
->exit_code
;
1311 if (unlikely(!exit_code
) ||
1312 unlikely(p
->state
& TASK_TRACED
))
1314 return wait_noreap_copyout(p
, pid
, uid
,
1315 why
, (exit_code
<< 8) | 0x7f,
1319 write_lock_irq(&tasklist_lock
);
1322 * This uses xchg to be atomic with the thread resuming and setting
1323 * it. It must also be done with the write lock held to prevent a
1324 * race with the EXIT_ZOMBIE case.
1326 exit_code
= xchg(&p
->exit_code
, 0);
1327 if (unlikely(p
->exit_state
)) {
1329 * The task resumed and then died. Let the next iteration
1330 * catch it in EXIT_ZOMBIE. Note that exit_code might
1331 * already be zero here if it resumed and did _exit(0).
1332 * The task itself is dead and won't touch exit_code again;
1333 * other processors in this function are locked out.
1335 p
->exit_code
= exit_code
;
1338 if (unlikely(exit_code
== 0)) {
1340 * Another thread in this function got to it first, or it
1341 * resumed, or it resumed and then died.
1343 write_unlock_irq(&tasklist_lock
);
1347 * We are returning to the wait loop without having successfully
1348 * removed the process and having released the lock. We cannot
1349 * continue, since the "p" task pointer is potentially stale.
1351 * Return -EAGAIN, and do_wait() will restart the loop from the
1352 * beginning. Do _not_ re-acquire the lock.
1357 /* move to end of parent's list to avoid starvation */
1361 write_unlock_irq(&tasklist_lock
);
1363 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1364 if (!retval
&& stat_addr
)
1365 retval
= put_user((exit_code
<< 8) | 0x7f, stat_addr
);
1366 if (!retval
&& infop
)
1367 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1368 if (!retval
&& infop
)
1369 retval
= put_user(0, &infop
->si_errno
);
1370 if (!retval
&& infop
)
1371 retval
= put_user((short)((p
->ptrace
& PT_PTRACED
)
1372 ? CLD_TRAPPED
: CLD_STOPPED
),
1374 if (!retval
&& infop
)
1375 retval
= put_user(exit_code
, &infop
->si_status
);
1376 if (!retval
&& infop
)
1377 retval
= put_user(p
->pid
, &infop
->si_pid
);
1378 if (!retval
&& infop
)
1379 retval
= put_user(p
->uid
, &infop
->si_uid
);
1389 * Handle do_wait work for one task in a live, non-stopped state.
1390 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1391 * the lock and this task is uninteresting. If we return nonzero, we have
1392 * released the lock and the system call should return.
1394 static int wait_task_continued(struct task_struct
*p
, int noreap
,
1395 struct siginfo __user
*infop
,
1396 int __user
*stat_addr
, struct rusage __user
*ru
)
1402 if (unlikely(!p
->signal
))
1405 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1408 spin_lock_irq(&p
->sighand
->siglock
);
1409 /* Re-check with the lock held. */
1410 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1411 spin_unlock_irq(&p
->sighand
->siglock
);
1415 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1416 spin_unlock_irq(&p
->sighand
->siglock
);
1421 read_unlock(&tasklist_lock
);
1424 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1426 if (!retval
&& stat_addr
)
1427 retval
= put_user(0xffff, stat_addr
);
1431 retval
= wait_noreap_copyout(p
, pid
, uid
,
1432 CLD_CONTINUED
, SIGCONT
,
1434 BUG_ON(retval
== 0);
1441 static inline int my_ptrace_child(struct task_struct
*p
)
1443 if (!(p
->ptrace
& PT_PTRACED
))
1445 if (!(p
->ptrace
& PT_ATTACHED
))
1448 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1449 * we are the attacher. If we are the real parent, this is a race
1450 * inside ptrace_attach. It is waiting for the tasklist_lock,
1451 * which we have to switch the parent links, but has already set
1452 * the flags in p->ptrace.
1454 return (p
->parent
!= p
->real_parent
);
1457 static long do_wait(pid_t pid
, int options
, struct siginfo __user
*infop
,
1458 int __user
*stat_addr
, struct rusage __user
*ru
)
1460 DECLARE_WAITQUEUE(wait
, current
);
1461 struct task_struct
*tsk
;
1463 int allowed
, denied
;
1465 add_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1468 * We will set this flag if we see any child that might later
1469 * match our criteria, even if we are not able to reap it yet.
1472 allowed
= denied
= 0;
1473 current
->state
= TASK_INTERRUPTIBLE
;
1474 read_lock(&tasklist_lock
);
1477 struct task_struct
*p
;
1478 struct list_head
*_p
;
1481 list_for_each(_p
,&tsk
->children
) {
1482 p
= list_entry(_p
, struct task_struct
, sibling
);
1484 ret
= eligible_child(pid
, options
, p
);
1488 if (unlikely(ret
< 0)) {
1497 * When we hit the race with PTRACE_ATTACH,
1498 * we will not report this child. But the
1499 * race means it has not yet been moved to
1500 * our ptrace_children list, so we need to
1501 * set the flag here to avoid a spurious ECHILD
1502 * when the race happens with the only child.
1505 if (!my_ptrace_child(p
))
1510 * It's stopped now, so it might later
1511 * continue, exit, or stop again.
1514 if (!(options
& WUNTRACED
) &&
1515 !my_ptrace_child(p
))
1517 retval
= wait_task_stopped(p
, ret
== 2,
1518 (options
& WNOWAIT
),
1521 if (retval
== -EAGAIN
)
1523 if (retval
!= 0) /* He released the lock. */
1528 if (p
->exit_state
== EXIT_DEAD
)
1530 // case EXIT_ZOMBIE:
1531 if (p
->exit_state
== EXIT_ZOMBIE
) {
1533 * Eligible but we cannot release
1537 goto check_continued
;
1538 if (!likely(options
& WEXITED
))
1540 retval
= wait_task_zombie(
1541 p
, (options
& WNOWAIT
),
1542 infop
, stat_addr
, ru
);
1543 /* He released the lock. */
1550 * It's running now, so it might later
1551 * exit, stop, or stop and then continue.
1554 if (!unlikely(options
& WCONTINUED
))
1556 retval
= wait_task_continued(
1557 p
, (options
& WNOWAIT
),
1558 infop
, stat_addr
, ru
);
1559 if (retval
!= 0) /* He released the lock. */
1565 list_for_each(_p
, &tsk
->ptrace_children
) {
1566 p
= list_entry(_p
, struct task_struct
,
1568 if (!eligible_child(pid
, options
, p
))
1574 if (options
& __WNOTHREAD
)
1576 tsk
= next_thread(tsk
);
1577 BUG_ON(tsk
->signal
!= current
->signal
);
1578 } while (tsk
!= current
);
1580 read_unlock(&tasklist_lock
);
1583 if (options
& WNOHANG
)
1585 retval
= -ERESTARTSYS
;
1586 if (signal_pending(current
))
1592 if (unlikely(denied
) && !allowed
)
1595 current
->state
= TASK_RUNNING
;
1596 remove_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1602 * For a WNOHANG return, clear out all the fields
1603 * we would set so the user can easily tell the
1607 retval
= put_user(0, &infop
->si_signo
);
1609 retval
= put_user(0, &infop
->si_errno
);
1611 retval
= put_user(0, &infop
->si_code
);
1613 retval
= put_user(0, &infop
->si_pid
);
1615 retval
= put_user(0, &infop
->si_uid
);
1617 retval
= put_user(0, &infop
->si_status
);
1623 asmlinkage
long sys_waitid(int which
, pid_t pid
,
1624 struct siginfo __user
*infop
, int options
,
1625 struct rusage __user
*ru
)
1629 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
))
1631 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1651 ret
= do_wait(pid
, options
, infop
, NULL
, ru
);
1653 /* avoid REGPARM breakage on x86: */
1654 prevent_tail_call(ret
);
1658 asmlinkage
long sys_wait4(pid_t pid
, int __user
*stat_addr
,
1659 int options
, struct rusage __user
*ru
)
1663 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1664 __WNOTHREAD
|__WCLONE
|__WALL
))
1666 ret
= do_wait(pid
, options
| WEXITED
, NULL
, stat_addr
, ru
);
1668 /* avoid REGPARM breakage on x86: */
1669 prevent_tail_call(ret
);
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 asmlinkage
long sys_waitpid(pid_t pid
, int __user
*stat_addr
, int options
)
1681 return sys_wait4(pid
, stat_addr
, options
, NULL
);