4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/smp_lock.h>
11 #include <linux/module.h>
12 #include <linux/capability.h>
13 #include <linux/completion.h>
14 #include <linux/personality.h>
15 #include <linux/tty.h>
16 #include <linux/mnt_namespace.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/binfmts.h>
24 #include <linux/nsproxy.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/ptrace.h>
27 #include <linux/profile.h>
28 #include <linux/mount.h>
29 #include <linux/proc_fs.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>
46 #include <asm/uaccess.h>
47 #include <asm/unistd.h>
48 #include <asm/pgtable.h>
49 #include <asm/mmu_context.h>
51 extern void sem_exit (void);
53 static void exit_mm(struct task_struct
* tsk
);
55 static void __unhash_process(struct task_struct
*p
)
58 detach_pid(p
, PIDTYPE_PID
);
59 if (thread_group_leader(p
)) {
60 detach_pid(p
, PIDTYPE_PGID
);
61 detach_pid(p
, PIDTYPE_SID
);
63 list_del_rcu(&p
->tasks
);
64 __get_cpu_var(process_counts
)--;
66 list_del_rcu(&p
->thread_group
);
71 * This function expects the tasklist_lock write-locked.
73 static void __exit_signal(struct task_struct
*tsk
)
75 struct signal_struct
*sig
= tsk
->signal
;
76 struct sighand_struct
*sighand
;
79 BUG_ON(!atomic_read(&sig
->count
));
82 sighand
= rcu_dereference(tsk
->sighand
);
83 spin_lock(&sighand
->siglock
);
85 posix_cpu_timers_exit(tsk
);
86 if (atomic_dec_and_test(&sig
->count
))
87 posix_cpu_timers_exit_group(tsk
);
90 * If there is any task waiting for the group exit
93 if (sig
->group_exit_task
&& atomic_read(&sig
->count
) == sig
->notify_count
) {
94 wake_up_process(sig
->group_exit_task
);
95 sig
->group_exit_task
= NULL
;
97 if (tsk
== sig
->curr_target
)
98 sig
->curr_target
= next_thread(tsk
);
100 * Accumulate here the counters for all threads but the
101 * group leader as they die, so they can be added into
102 * the process-wide totals when those are taken.
103 * The group leader stays around as a zombie as long
104 * as there are other threads. When it gets reaped,
105 * the exit.c code will add its counts into these totals.
106 * We won't ever get here for the group leader, since it
107 * will have been the last reference on the signal_struct.
109 sig
->utime
= cputime_add(sig
->utime
, tsk
->utime
);
110 sig
->stime
= cputime_add(sig
->stime
, tsk
->stime
);
111 sig
->min_flt
+= tsk
->min_flt
;
112 sig
->maj_flt
+= tsk
->maj_flt
;
113 sig
->nvcsw
+= tsk
->nvcsw
;
114 sig
->nivcsw
+= tsk
->nivcsw
;
115 sig
->sched_time
+= tsk
->sched_time
;
116 sig
= NULL
; /* Marker for below. */
119 __unhash_process(tsk
);
123 spin_unlock(&sighand
->siglock
);
126 __cleanup_sighand(sighand
);
127 clear_tsk_thread_flag(tsk
,TIF_SIGPENDING
);
128 flush_sigqueue(&tsk
->pending
);
130 flush_sigqueue(&sig
->shared_pending
);
131 taskstats_tgid_free(sig
);
132 __cleanup_signal(sig
);
136 static void delayed_put_task_struct(struct rcu_head
*rhp
)
138 put_task_struct(container_of(rhp
, struct task_struct
, rcu
));
141 void release_task(struct task_struct
* p
)
143 struct task_struct
*leader
;
146 atomic_dec(&p
->user
->processes
);
147 write_lock_irq(&tasklist_lock
);
149 BUG_ON(!list_empty(&p
->ptrace_list
) || !list_empty(&p
->ptrace_children
));
153 * If we are the last non-leader member of the thread
154 * group, and the leader is zombie, then notify the
155 * group leader's parent process. (if it wants notification.)
158 leader
= p
->group_leader
;
159 if (leader
!= p
&& thread_group_empty(leader
) && leader
->exit_state
== EXIT_ZOMBIE
) {
160 BUG_ON(leader
->exit_signal
== -1);
161 do_notify_parent(leader
, leader
->exit_signal
);
163 * If we were the last child thread and the leader has
164 * exited already, and the leader's parent ignores SIGCHLD,
165 * then we are the one who should release the leader.
167 * do_notify_parent() will have marked it self-reaping in
170 zap_leader
= (leader
->exit_signal
== -1);
174 write_unlock_irq(&tasklist_lock
);
177 call_rcu(&p
->rcu
, delayed_put_task_struct
);
180 if (unlikely(zap_leader
))
185 * This checks not only the pgrp, but falls back on the pid if no
186 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
189 int session_of_pgrp(int pgrp
)
191 struct task_struct
*p
;
194 read_lock(&tasklist_lock
);
196 p
= find_task_by_pid_type(PIDTYPE_PGID
, pgrp
);
198 p
= find_task_by_pid(pgrp
);
200 sid
= process_session(p
);
202 read_unlock(&tasklist_lock
);
208 * Determine if a process group is "orphaned", according to the POSIX
209 * definition in 2.2.2.52. Orphaned process groups are not to be affected
210 * by terminal-generated stop signals. Newly orphaned process groups are
211 * to receive a SIGHUP and a SIGCONT.
213 * "I ask you, have you ever known what it is to be an orphan?"
215 static int will_become_orphaned_pgrp(int pgrp
, struct task_struct
*ignored_task
)
217 struct task_struct
*p
;
220 do_each_task_pid(pgrp
, PIDTYPE_PGID
, p
) {
221 if (p
== ignored_task
223 || is_init(p
->real_parent
))
225 if (process_group(p
->real_parent
) != pgrp
&&
226 process_session(p
->real_parent
) == process_session(p
)) {
230 } while_each_task_pid(pgrp
, PIDTYPE_PGID
, p
);
231 return ret
; /* (sighing) "Often!" */
234 int is_orphaned_pgrp(int pgrp
)
238 read_lock(&tasklist_lock
);
239 retval
= will_become_orphaned_pgrp(pgrp
, NULL
);
240 read_unlock(&tasklist_lock
);
245 static int has_stopped_jobs(int pgrp
)
248 struct task_struct
*p
;
250 do_each_task_pid(pgrp
, PIDTYPE_PGID
, p
) {
251 if (p
->state
!= TASK_STOPPED
)
255 } while_each_task_pid(pgrp
, PIDTYPE_PGID
, p
);
260 * reparent_to_init - Reparent the calling kernel thread to the init task of the pid space that the thread belongs to.
262 * If a kernel thread is launched as a result of a system call, or if
263 * it ever exits, it should generally reparent itself to init so that
264 * it is correctly cleaned up on exit.
266 * The various task state such as scheduling policy and priority may have
267 * been inherited from a user process, so we reset them to sane values here.
269 * NOTE that reparent_to_init() gives the caller full capabilities.
271 static void reparent_to_init(void)
273 write_lock_irq(&tasklist_lock
);
275 ptrace_unlink(current
);
276 /* Reparent to init */
277 remove_parent(current
);
278 current
->parent
= child_reaper(current
);
279 current
->real_parent
= child_reaper(current
);
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 sigaddset(¤t
->blocked
, sig
);
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
);
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
);
442 struct files_struct
*get_files_struct(struct task_struct
*task
)
444 struct files_struct
*files
;
449 atomic_inc(&files
->count
);
455 void fastcall
put_files_struct(struct files_struct
*files
)
459 if (atomic_dec_and_test(&files
->count
)) {
462 * Free the fd and fdset arrays if we expanded them.
463 * If the fdtable was embedded, pass files for freeing
464 * at the end of the RCU grace period. Otherwise,
465 * you can free files immediately.
467 fdt
= files_fdtable(files
);
468 if (fdt
!= &files
->fdtab
)
469 kmem_cache_free(files_cachep
, files
);
474 EXPORT_SYMBOL(put_files_struct
);
476 void reset_files_struct(struct task_struct
*tsk
, struct files_struct
*files
)
478 struct files_struct
*old
;
484 put_files_struct(old
);
486 EXPORT_SYMBOL(reset_files_struct
);
488 static inline void __exit_files(struct task_struct
*tsk
)
490 struct files_struct
* files
= tsk
->files
;
496 put_files_struct(files
);
500 void exit_files(struct task_struct
*tsk
)
505 static inline void __put_fs_struct(struct fs_struct
*fs
)
507 /* No need to hold fs->lock if we are killing it */
508 if (atomic_dec_and_test(&fs
->count
)) {
515 mntput(fs
->altrootmnt
);
517 kmem_cache_free(fs_cachep
, fs
);
521 void put_fs_struct(struct fs_struct
*fs
)
526 static inline void __exit_fs(struct task_struct
*tsk
)
528 struct fs_struct
* fs
= tsk
->fs
;
538 void exit_fs(struct task_struct
*tsk
)
543 EXPORT_SYMBOL_GPL(exit_fs
);
546 * Turn us into a lazy TLB process if we
549 static void exit_mm(struct task_struct
* tsk
)
551 struct mm_struct
*mm
= tsk
->mm
;
557 * Serialize with any possible pending coredump.
558 * We must hold mmap_sem around checking core_waiters
559 * and clearing tsk->mm. The core-inducing thread
560 * will increment core_waiters for each thread in the
561 * group with ->mm != NULL.
563 down_read(&mm
->mmap_sem
);
564 if (mm
->core_waiters
) {
565 up_read(&mm
->mmap_sem
);
566 down_write(&mm
->mmap_sem
);
567 if (!--mm
->core_waiters
)
568 complete(mm
->core_startup_done
);
569 up_write(&mm
->mmap_sem
);
571 wait_for_completion(&mm
->core_done
);
572 down_read(&mm
->mmap_sem
);
574 atomic_inc(&mm
->mm_count
);
575 BUG_ON(mm
!= tsk
->active_mm
);
576 /* more a memory barrier than a real lock */
579 up_read(&mm
->mmap_sem
);
580 enter_lazy_tlb(mm
, current
);
586 choose_new_parent(struct task_struct
*p
, struct task_struct
*reaper
)
589 * Make sure we're not reparenting to ourselves and that
590 * the parent is not a zombie.
592 BUG_ON(p
== reaper
|| reaper
->exit_state
);
593 p
->real_parent
= reaper
;
597 reparent_thread(struct task_struct
*p
, struct task_struct
*father
, int traced
)
599 if (p
->pdeath_signal
)
600 /* We already hold the tasklist_lock here. */
601 group_send_sig_info(p
->pdeath_signal
, SEND_SIG_NOINFO
, p
);
603 /* Move the child from its dying parent to the new one. */
604 if (unlikely(traced
)) {
605 /* Preserve ptrace links if someone else is tracing this child. */
606 list_del_init(&p
->ptrace_list
);
607 if (p
->parent
!= p
->real_parent
)
608 list_add(&p
->ptrace_list
, &p
->real_parent
->ptrace_children
);
610 /* If this child is being traced, then we're the one tracing it
611 * anyway, so let go of it.
615 p
->parent
= p
->real_parent
;
618 if (p
->state
== TASK_TRACED
) {
620 * If it was at a trace stop, turn it into
621 * a normal stop since it's no longer being
628 /* If this is a threaded reparent there is no need to
629 * notify anyone anything has happened.
631 if (p
->real_parent
->group_leader
== father
->group_leader
)
634 /* We don't want people slaying init. */
635 if (p
->exit_signal
!= -1)
636 p
->exit_signal
= SIGCHLD
;
638 /* If we'd notified the old parent about this child's death,
639 * also notify the new parent.
641 if (!traced
&& p
->exit_state
== EXIT_ZOMBIE
&&
642 p
->exit_signal
!= -1 && thread_group_empty(p
))
643 do_notify_parent(p
, p
->exit_signal
);
646 * process group orphan check
647 * Case ii: Our child is in a different pgrp
648 * than we are, and it was the only connection
649 * outside, so the child pgrp is now orphaned.
651 if ((process_group(p
) != process_group(father
)) &&
652 (process_session(p
) == process_session(father
))) {
653 int pgrp
= process_group(p
);
655 if (will_become_orphaned_pgrp(pgrp
, NULL
) &&
656 has_stopped_jobs(pgrp
)) {
657 __kill_pg_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
658 __kill_pg_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
664 * When we die, we re-parent all our children.
665 * Try to give them to another thread in our thread
666 * group, and if no such member exists, give it to
667 * the child reaper process (ie "init") in our pid
671 forget_original_parent(struct task_struct
*father
, struct list_head
*to_release
)
673 struct task_struct
*p
, *reaper
= father
;
674 struct list_head
*_p
, *_n
;
677 reaper
= next_thread(reaper
);
678 if (reaper
== father
) {
679 reaper
= child_reaper(father
);
682 } while (reaper
->exit_state
);
685 * There are only two places where our children can be:
687 * - in our child list
688 * - in our ptraced child list
690 * Search them and reparent children.
692 list_for_each_safe(_p
, _n
, &father
->children
) {
694 p
= list_entry(_p
, struct task_struct
, sibling
);
698 /* if father isn't the real parent, then ptrace must be enabled */
699 BUG_ON(father
!= p
->real_parent
&& !ptrace
);
701 if (father
== p
->real_parent
) {
702 /* reparent with a reaper, real father it's us */
703 choose_new_parent(p
, reaper
);
704 reparent_thread(p
, father
, 0);
706 /* reparent ptraced task to its real parent */
708 if (p
->exit_state
== EXIT_ZOMBIE
&& p
->exit_signal
!= -1 &&
709 thread_group_empty(p
))
710 do_notify_parent(p
, p
->exit_signal
);
714 * if the ptraced child is a zombie with exit_signal == -1
715 * we must collect it before we exit, or it will remain
716 * zombie forever since we prevented it from self-reap itself
717 * while it was being traced by us, to be able to see it in wait4.
719 if (unlikely(ptrace
&& p
->exit_state
== EXIT_ZOMBIE
&& p
->exit_signal
== -1))
720 list_add(&p
->ptrace_list
, to_release
);
722 list_for_each_safe(_p
, _n
, &father
->ptrace_children
) {
723 p
= list_entry(_p
, struct task_struct
, ptrace_list
);
724 choose_new_parent(p
, reaper
);
725 reparent_thread(p
, father
, 1);
730 * Send signals to all our closest relatives so that they know
731 * to properly mourn us..
733 static void exit_notify(struct task_struct
*tsk
)
736 struct task_struct
*t
;
737 struct list_head ptrace_dead
, *_p
, *_n
;
739 if (signal_pending(tsk
) && !(tsk
->signal
->flags
& SIGNAL_GROUP_EXIT
)
740 && !thread_group_empty(tsk
)) {
742 * This occurs when there was a race between our exit
743 * syscall and a group signal choosing us as the one to
744 * wake up. It could be that we are the only thread
745 * alerted to check for pending signals, but another thread
746 * should be woken now to take the signal since we will not.
747 * Now we'll wake all the threads in the group just to make
748 * sure someone gets all the pending signals.
750 read_lock(&tasklist_lock
);
751 spin_lock_irq(&tsk
->sighand
->siglock
);
752 for (t
= next_thread(tsk
); t
!= tsk
; t
= next_thread(t
))
753 if (!signal_pending(t
) && !(t
->flags
& PF_EXITING
)) {
754 recalc_sigpending_tsk(t
);
755 if (signal_pending(t
))
756 signal_wake_up(t
, 0);
758 spin_unlock_irq(&tsk
->sighand
->siglock
);
759 read_unlock(&tasklist_lock
);
762 write_lock_irq(&tasklist_lock
);
765 * This does two things:
767 * A. Make init inherit all the child processes
768 * B. Check to see if any process groups have become orphaned
769 * as a result of our exiting, and if they have any stopped
770 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
773 INIT_LIST_HEAD(&ptrace_dead
);
774 forget_original_parent(tsk
, &ptrace_dead
);
775 BUG_ON(!list_empty(&tsk
->children
));
776 BUG_ON(!list_empty(&tsk
->ptrace_children
));
779 * Check to see if any process groups have become orphaned
780 * as a result of our exiting, and if they have any stopped
781 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
783 * Case i: Our father is in a different pgrp than we are
784 * and we were the only connection outside, so our pgrp
785 * is about to become orphaned.
788 t
= tsk
->real_parent
;
790 if ((process_group(t
) != process_group(tsk
)) &&
791 (process_session(t
) == process_session(tsk
)) &&
792 will_become_orphaned_pgrp(process_group(tsk
), tsk
) &&
793 has_stopped_jobs(process_group(tsk
))) {
794 __kill_pg_info(SIGHUP
, SEND_SIG_PRIV
, process_group(tsk
));
795 __kill_pg_info(SIGCONT
, SEND_SIG_PRIV
, process_group(tsk
));
798 /* Let father know we died
800 * Thread signals are configurable, but you aren't going to use
801 * that to send signals to arbitary processes.
802 * That stops right now.
804 * If the parent exec id doesn't match the exec id we saved
805 * when we started then we know the parent has changed security
808 * If our self_exec id doesn't match our parent_exec_id then
809 * we have changed execution domain as these two values started
810 * the same after a fork.
814 if (tsk
->exit_signal
!= SIGCHLD
&& tsk
->exit_signal
!= -1 &&
815 ( tsk
->parent_exec_id
!= t
->self_exec_id
||
816 tsk
->self_exec_id
!= tsk
->parent_exec_id
)
817 && !capable(CAP_KILL
))
818 tsk
->exit_signal
= SIGCHLD
;
821 /* If something other than our normal parent is ptracing us, then
822 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
823 * only has special meaning to our real parent.
825 if (tsk
->exit_signal
!= -1 && thread_group_empty(tsk
)) {
826 int signal
= tsk
->parent
== tsk
->real_parent
? tsk
->exit_signal
: SIGCHLD
;
827 do_notify_parent(tsk
, signal
);
828 } else if (tsk
->ptrace
) {
829 do_notify_parent(tsk
, SIGCHLD
);
833 if (tsk
->exit_signal
== -1 &&
834 (likely(tsk
->ptrace
== 0) ||
835 unlikely(tsk
->parent
->signal
->flags
& SIGNAL_GROUP_EXIT
)))
837 tsk
->exit_state
= state
;
839 write_unlock_irq(&tasklist_lock
);
841 list_for_each_safe(_p
, _n
, &ptrace_dead
) {
843 t
= list_entry(_p
, struct task_struct
, ptrace_list
);
847 /* If the process is dead, release it - nobody will wait for it */
848 if (state
== EXIT_DEAD
)
852 fastcall NORET_TYPE
void do_exit(long code
)
854 struct task_struct
*tsk
= current
;
857 profile_task_exit(tsk
);
859 WARN_ON(atomic_read(&tsk
->fs_excl
));
861 if (unlikely(in_interrupt()))
862 panic("Aiee, killing interrupt handler!");
863 if (unlikely(!tsk
->pid
))
864 panic("Attempted to kill the idle task!");
865 if (unlikely(tsk
== child_reaper(tsk
))) {
866 if (tsk
->nsproxy
->pid_ns
!= &init_pid_ns
)
867 tsk
->nsproxy
->pid_ns
->child_reaper
= init_pid_ns
.child_reaper
;
869 panic("Attempted to kill init!");
873 if (unlikely(current
->ptrace
& PT_TRACE_EXIT
)) {
874 current
->ptrace_message
= code
;
875 ptrace_notify((PTRACE_EVENT_EXIT
<< 8) | SIGTRAP
);
879 * We're taking recursive faults here in do_exit. Safest is to just
880 * leave this task alone and wait for reboot.
882 if (unlikely(tsk
->flags
& PF_EXITING
)) {
884 "Fixing recursive fault but reboot is needed!\n");
887 set_current_state(TASK_UNINTERRUPTIBLE
);
891 tsk
->flags
|= PF_EXITING
;
893 if (unlikely(in_atomic()))
894 printk(KERN_INFO
"note: %s[%d] exited with preempt_count %d\n",
895 current
->comm
, current
->pid
,
898 acct_update_integrals(tsk
);
900 update_hiwater_rss(tsk
->mm
);
901 update_hiwater_vm(tsk
->mm
);
903 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
905 hrtimer_cancel(&tsk
->signal
->real_timer
);
906 exit_itimers(tsk
->signal
);
908 acct_collect(code
, group_dead
);
909 if (unlikely(tsk
->robust_list
))
910 exit_robust_list(tsk
);
911 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
912 if (unlikely(tsk
->compat_robust_list
))
913 compat_exit_robust_list(tsk
);
915 if (unlikely(tsk
->audit_context
))
918 taskstats_exit(tsk
, group_dead
);
931 if (group_dead
&& tsk
->signal
->leader
)
932 disassociate_ctty(1);
934 module_put(task_thread_info(tsk
)->exec_domain
->module
);
936 module_put(tsk
->binfmt
->module
);
938 tsk
->exit_code
= code
;
939 proc_exit_connector(tsk
);
940 exit_task_namespaces(tsk
);
943 mpol_free(tsk
->mempolicy
);
944 tsk
->mempolicy
= NULL
;
947 * This must happen late, after the PID is not
950 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
951 exit_pi_state_list(tsk
);
952 if (unlikely(current
->pi_state_cache
))
953 kfree(current
->pi_state_cache
);
955 * Make sure we are holding no locks:
957 debug_check_no_locks_held(tsk
);
962 if (tsk
->splice_pipe
)
963 __free_pipe_info(tsk
->splice_pipe
);
966 /* causes final put_task_struct in finish_task_switch(). */
967 tsk
->state
= TASK_DEAD
;
971 /* Avoid "noreturn function does return". */
973 cpu_relax(); /* For when BUG is null */
976 EXPORT_SYMBOL_GPL(do_exit
);
978 NORET_TYPE
void complete_and_exit(struct completion
*comp
, long code
)
986 EXPORT_SYMBOL(complete_and_exit
);
988 asmlinkage
long sys_exit(int error_code
)
990 do_exit((error_code
&0xff)<<8);
994 * Take down every thread in the group. This is called by fatal signals
995 * as well as by sys_exit_group (below).
998 do_group_exit(int exit_code
)
1000 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
1002 if (current
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1003 exit_code
= current
->signal
->group_exit_code
;
1004 else if (!thread_group_empty(current
)) {
1005 struct signal_struct
*const sig
= current
->signal
;
1006 struct sighand_struct
*const sighand
= current
->sighand
;
1007 spin_lock_irq(&sighand
->siglock
);
1008 if (sig
->flags
& SIGNAL_GROUP_EXIT
)
1009 /* Another thread got here before we took the lock. */
1010 exit_code
= sig
->group_exit_code
;
1012 sig
->group_exit_code
= exit_code
;
1013 zap_other_threads(current
);
1015 spin_unlock_irq(&sighand
->siglock
);
1023 * this kills every thread in the thread group. Note that any externally
1024 * wait4()-ing process will get the correct exit code - even if this
1025 * thread is not the thread group leader.
1027 asmlinkage
void sys_exit_group(int error_code
)
1029 do_group_exit((error_code
& 0xff) << 8);
1032 static int eligible_child(pid_t pid
, int options
, struct task_struct
*p
)
1038 if (process_group(p
) != process_group(current
))
1040 } else if (pid
!= -1) {
1041 if (process_group(p
) != -pid
)
1046 * Do not consider detached threads that are
1049 if (p
->exit_signal
== -1 && !p
->ptrace
)
1052 /* Wait for all children (clone and not) if __WALL is set;
1053 * otherwise, wait for clone children *only* if __WCLONE is
1054 * set; otherwise, wait for non-clone children *only*. (Note:
1055 * A "clone" child here is one that reports to its parent
1056 * using a signal other than SIGCHLD.) */
1057 if (((p
->exit_signal
!= SIGCHLD
) ^ ((options
& __WCLONE
) != 0))
1058 && !(options
& __WALL
))
1061 * Do not consider thread group leaders that are
1062 * in a non-empty thread group:
1064 if (delay_group_leader(p
))
1067 if (security_task_wait(p
))
1073 static int wait_noreap_copyout(struct task_struct
*p
, pid_t pid
, uid_t uid
,
1074 int why
, int status
,
1075 struct siginfo __user
*infop
,
1076 struct rusage __user
*rusagep
)
1078 int retval
= rusagep
? getrusage(p
, RUSAGE_BOTH
, rusagep
) : 0;
1082 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1084 retval
= put_user(0, &infop
->si_errno
);
1086 retval
= put_user((short)why
, &infop
->si_code
);
1088 retval
= put_user(pid
, &infop
->si_pid
);
1090 retval
= put_user(uid
, &infop
->si_uid
);
1092 retval
= put_user(status
, &infop
->si_status
);
1099 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1100 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1101 * the lock and this task is uninteresting. If we return nonzero, we have
1102 * released the lock and the system call should return.
1104 static int wait_task_zombie(struct task_struct
*p
, int noreap
,
1105 struct siginfo __user
*infop
,
1106 int __user
*stat_addr
, struct rusage __user
*ru
)
1108 unsigned long state
;
1112 if (unlikely(noreap
)) {
1115 int exit_code
= p
->exit_code
;
1118 if (unlikely(p
->exit_state
!= EXIT_ZOMBIE
))
1120 if (unlikely(p
->exit_signal
== -1 && p
->ptrace
== 0))
1123 read_unlock(&tasklist_lock
);
1124 if ((exit_code
& 0x7f) == 0) {
1126 status
= exit_code
>> 8;
1128 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1129 status
= exit_code
& 0x7f;
1131 return wait_noreap_copyout(p
, pid
, uid
, why
,
1136 * Try to move the task's state to DEAD
1137 * only one thread is allowed to do this:
1139 state
= xchg(&p
->exit_state
, EXIT_DEAD
);
1140 if (state
!= EXIT_ZOMBIE
) {
1141 BUG_ON(state
!= EXIT_DEAD
);
1144 if (unlikely(p
->exit_signal
== -1 && p
->ptrace
== 0)) {
1146 * This can only happen in a race with a ptraced thread
1147 * dying on another processor.
1152 if (likely(p
->real_parent
== p
->parent
) && likely(p
->signal
)) {
1153 struct signal_struct
*psig
;
1154 struct signal_struct
*sig
;
1157 * The resource counters for the group leader are in its
1158 * own task_struct. Those for dead threads in the group
1159 * are in its signal_struct, as are those for the child
1160 * processes it has previously reaped. All these
1161 * accumulate in the parent's signal_struct c* fields.
1163 * We don't bother to take a lock here to protect these
1164 * p->signal fields, because they are only touched by
1165 * __exit_signal, which runs with tasklist_lock
1166 * write-locked anyway, and so is excluded here. We do
1167 * need to protect the access to p->parent->signal fields,
1168 * as other threads in the parent group can be right
1169 * here reaping other children at the same time.
1171 spin_lock_irq(&p
->parent
->sighand
->siglock
);
1172 psig
= p
->parent
->signal
;
1175 cputime_add(psig
->cutime
,
1176 cputime_add(p
->utime
,
1177 cputime_add(sig
->utime
,
1180 cputime_add(psig
->cstime
,
1181 cputime_add(p
->stime
,
1182 cputime_add(sig
->stime
,
1185 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1187 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1189 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1191 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1192 spin_unlock_irq(&p
->parent
->sighand
->siglock
);
1196 * Now we are sure this task is interesting, and no other
1197 * thread can reap it because we set its state to EXIT_DEAD.
1199 read_unlock(&tasklist_lock
);
1201 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1202 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1203 ? p
->signal
->group_exit_code
: p
->exit_code
;
1204 if (!retval
&& stat_addr
)
1205 retval
= put_user(status
, stat_addr
);
1206 if (!retval
&& infop
)
1207 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1208 if (!retval
&& infop
)
1209 retval
= put_user(0, &infop
->si_errno
);
1210 if (!retval
&& infop
) {
1213 if ((status
& 0x7f) == 0) {
1217 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1220 retval
= put_user((short)why
, &infop
->si_code
);
1222 retval
= put_user(status
, &infop
->si_status
);
1224 if (!retval
&& infop
)
1225 retval
= put_user(p
->pid
, &infop
->si_pid
);
1226 if (!retval
&& infop
)
1227 retval
= put_user(p
->uid
, &infop
->si_uid
);
1229 // TODO: is this safe?
1230 p
->exit_state
= EXIT_ZOMBIE
;
1234 if (p
->real_parent
!= p
->parent
) {
1235 write_lock_irq(&tasklist_lock
);
1236 /* Double-check with lock held. */
1237 if (p
->real_parent
!= p
->parent
) {
1239 // TODO: is this safe?
1240 p
->exit_state
= EXIT_ZOMBIE
;
1242 * If this is not a detached task, notify the parent.
1243 * If it's still not detached after that, don't release
1246 if (p
->exit_signal
!= -1) {
1247 do_notify_parent(p
, p
->exit_signal
);
1248 if (p
->exit_signal
!= -1)
1252 write_unlock_irq(&tasklist_lock
);
1261 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1262 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1263 * the lock and this task is uninteresting. If we return nonzero, we have
1264 * released the lock and the system call should return.
1266 static int wait_task_stopped(struct task_struct
*p
, int delayed_group_leader
,
1267 int noreap
, struct siginfo __user
*infop
,
1268 int __user
*stat_addr
, struct rusage __user
*ru
)
1270 int retval
, exit_code
;
1274 if (delayed_group_leader
&& !(p
->ptrace
& PT_PTRACED
) &&
1275 p
->signal
&& p
->signal
->group_stop_count
> 0)
1277 * A group stop is in progress and this is the group leader.
1278 * We won't report until all threads have stopped.
1283 * Now we are pretty sure this task is interesting.
1284 * Make sure it doesn't get reaped out from under us while we
1285 * give up the lock and then examine it below. We don't want to
1286 * keep holding onto the tasklist_lock while we call getrusage and
1287 * possibly take page faults for user memory.
1290 read_unlock(&tasklist_lock
);
1292 if (unlikely(noreap
)) {
1295 int why
= (p
->ptrace
& PT_PTRACED
) ? CLD_TRAPPED
: CLD_STOPPED
;
1297 exit_code
= p
->exit_code
;
1298 if (unlikely(!exit_code
) ||
1299 unlikely(p
->state
& TASK_TRACED
))
1301 return wait_noreap_copyout(p
, pid
, uid
,
1302 why
, (exit_code
<< 8) | 0x7f,
1306 write_lock_irq(&tasklist_lock
);
1309 * This uses xchg to be atomic with the thread resuming and setting
1310 * it. It must also be done with the write lock held to prevent a
1311 * race with the EXIT_ZOMBIE case.
1313 exit_code
= xchg(&p
->exit_code
, 0);
1314 if (unlikely(p
->exit_state
)) {
1316 * The task resumed and then died. Let the next iteration
1317 * catch it in EXIT_ZOMBIE. Note that exit_code might
1318 * already be zero here if it resumed and did _exit(0).
1319 * The task itself is dead and won't touch exit_code again;
1320 * other processors in this function are locked out.
1322 p
->exit_code
= exit_code
;
1325 if (unlikely(exit_code
== 0)) {
1327 * Another thread in this function got to it first, or it
1328 * resumed, or it resumed and then died.
1330 write_unlock_irq(&tasklist_lock
);
1334 * We are returning to the wait loop without having successfully
1335 * removed the process and having released the lock. We cannot
1336 * continue, since the "p" task pointer is potentially stale.
1338 * Return -EAGAIN, and do_wait() will restart the loop from the
1339 * beginning. Do _not_ re-acquire the lock.
1344 /* move to end of parent's list to avoid starvation */
1348 write_unlock_irq(&tasklist_lock
);
1350 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1351 if (!retval
&& stat_addr
)
1352 retval
= put_user((exit_code
<< 8) | 0x7f, stat_addr
);
1353 if (!retval
&& infop
)
1354 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1355 if (!retval
&& infop
)
1356 retval
= put_user(0, &infop
->si_errno
);
1357 if (!retval
&& infop
)
1358 retval
= put_user((short)((p
->ptrace
& PT_PTRACED
)
1359 ? CLD_TRAPPED
: CLD_STOPPED
),
1361 if (!retval
&& infop
)
1362 retval
= put_user(exit_code
, &infop
->si_status
);
1363 if (!retval
&& infop
)
1364 retval
= put_user(p
->pid
, &infop
->si_pid
);
1365 if (!retval
&& infop
)
1366 retval
= put_user(p
->uid
, &infop
->si_uid
);
1376 * Handle do_wait work for one task in a live, non-stopped state.
1377 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1378 * the lock and this task is uninteresting. If we return nonzero, we have
1379 * released the lock and the system call should return.
1381 static int wait_task_continued(struct task_struct
*p
, int noreap
,
1382 struct siginfo __user
*infop
,
1383 int __user
*stat_addr
, struct rusage __user
*ru
)
1389 if (unlikely(!p
->signal
))
1392 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1395 spin_lock_irq(&p
->sighand
->siglock
);
1396 /* Re-check with the lock held. */
1397 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1398 spin_unlock_irq(&p
->sighand
->siglock
);
1402 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1403 spin_unlock_irq(&p
->sighand
->siglock
);
1408 read_unlock(&tasklist_lock
);
1411 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1413 if (!retval
&& stat_addr
)
1414 retval
= put_user(0xffff, stat_addr
);
1418 retval
= wait_noreap_copyout(p
, pid
, uid
,
1419 CLD_CONTINUED
, SIGCONT
,
1421 BUG_ON(retval
== 0);
1428 static inline int my_ptrace_child(struct task_struct
*p
)
1430 if (!(p
->ptrace
& PT_PTRACED
))
1432 if (!(p
->ptrace
& PT_ATTACHED
))
1435 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1436 * we are the attacher. If we are the real parent, this is a race
1437 * inside ptrace_attach. It is waiting for the tasklist_lock,
1438 * which we have to switch the parent links, but has already set
1439 * the flags in p->ptrace.
1441 return (p
->parent
!= p
->real_parent
);
1444 static long do_wait(pid_t pid
, int options
, struct siginfo __user
*infop
,
1445 int __user
*stat_addr
, struct rusage __user
*ru
)
1447 DECLARE_WAITQUEUE(wait
, current
);
1448 struct task_struct
*tsk
;
1451 add_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1454 * We will set this flag if we see any child that might later
1455 * match our criteria, even if we are not able to reap it yet.
1458 current
->state
= TASK_INTERRUPTIBLE
;
1459 read_lock(&tasklist_lock
);
1462 struct task_struct
*p
;
1463 struct list_head
*_p
;
1466 list_for_each(_p
,&tsk
->children
) {
1467 p
= list_entry(_p
, struct task_struct
, sibling
);
1469 ret
= eligible_child(pid
, options
, p
);
1476 * When we hit the race with PTRACE_ATTACH,
1477 * we will not report this child. But the
1478 * race means it has not yet been moved to
1479 * our ptrace_children list, so we need to
1480 * set the flag here to avoid a spurious ECHILD
1481 * when the race happens with the only child.
1484 if (!my_ptrace_child(p
))
1489 * It's stopped now, so it might later
1490 * continue, exit, or stop again.
1493 if (!(options
& WUNTRACED
) &&
1494 !my_ptrace_child(p
))
1496 retval
= wait_task_stopped(p
, ret
== 2,
1497 (options
& WNOWAIT
),
1500 if (retval
== -EAGAIN
)
1502 if (retval
!= 0) /* He released the lock. */
1507 if (p
->exit_state
== EXIT_DEAD
)
1509 // case EXIT_ZOMBIE:
1510 if (p
->exit_state
== EXIT_ZOMBIE
) {
1512 * Eligible but we cannot release
1516 goto check_continued
;
1517 if (!likely(options
& WEXITED
))
1519 retval
= wait_task_zombie(
1520 p
, (options
& WNOWAIT
),
1521 infop
, stat_addr
, ru
);
1522 /* He released the lock. */
1529 * It's running now, so it might later
1530 * exit, stop, or stop and then continue.
1533 if (!unlikely(options
& WCONTINUED
))
1535 retval
= wait_task_continued(
1536 p
, (options
& WNOWAIT
),
1537 infop
, stat_addr
, ru
);
1538 if (retval
!= 0) /* He released the lock. */
1544 list_for_each(_p
, &tsk
->ptrace_children
) {
1545 p
= list_entry(_p
, struct task_struct
,
1547 if (!eligible_child(pid
, options
, p
))
1553 if (options
& __WNOTHREAD
)
1555 tsk
= next_thread(tsk
);
1556 BUG_ON(tsk
->signal
!= current
->signal
);
1557 } while (tsk
!= current
);
1559 read_unlock(&tasklist_lock
);
1562 if (options
& WNOHANG
)
1564 retval
= -ERESTARTSYS
;
1565 if (signal_pending(current
))
1572 current
->state
= TASK_RUNNING
;
1573 remove_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1579 * For a WNOHANG return, clear out all the fields
1580 * we would set so the user can easily tell the
1584 retval
= put_user(0, &infop
->si_signo
);
1586 retval
= put_user(0, &infop
->si_errno
);
1588 retval
= put_user(0, &infop
->si_code
);
1590 retval
= put_user(0, &infop
->si_pid
);
1592 retval
= put_user(0, &infop
->si_uid
);
1594 retval
= put_user(0, &infop
->si_status
);
1600 asmlinkage
long sys_waitid(int which
, pid_t pid
,
1601 struct siginfo __user
*infop
, int options
,
1602 struct rusage __user
*ru
)
1606 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
))
1608 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1628 ret
= do_wait(pid
, options
, infop
, NULL
, ru
);
1630 /* avoid REGPARM breakage on x86: */
1631 prevent_tail_call(ret
);
1635 asmlinkage
long sys_wait4(pid_t pid
, int __user
*stat_addr
,
1636 int options
, struct rusage __user
*ru
)
1640 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1641 __WNOTHREAD
|__WCLONE
|__WALL
))
1643 ret
= do_wait(pid
, options
| WEXITED
, NULL
, stat_addr
, ru
);
1645 /* avoid REGPARM breakage on x86: */
1646 prevent_tail_call(ret
);
1650 #ifdef __ARCH_WANT_SYS_WAITPID
1653 * sys_waitpid() remains for compatibility. waitpid() should be
1654 * implemented by calling sys_wait4() from libc.a.
1656 asmlinkage
long sys_waitpid(pid_t pid
, int __user
*stat_addr
, int options
)
1658 return sys_wait4(pid
, stat_addr
, options
, NULL
);