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
261 * of the pid space that the thread belongs to.
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 init so that
265 * it 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_init() gives the caller full capabilities.
272 static void reparent_to_init(void)
274 write_lock_irq(&tasklist_lock
);
276 ptrace_unlink(current
);
277 /* Reparent to init */
278 remove_parent(current
);
279 current
->parent
= child_reaper(current
);
280 current
->real_parent
= child_reaper(current
);
283 /* Set the exit signal to SIGCHLD so we signal init on exit */
284 current
->exit_signal
= SIGCHLD
;
286 if (!has_rt_policy(current
) && (task_nice(current
) < 0))
287 set_user_nice(current
, 0);
291 security_task_reparent_to_init(current
);
292 memcpy(current
->signal
->rlim
, init_task
.signal
->rlim
,
293 sizeof(current
->signal
->rlim
));
294 atomic_inc(&(INIT_USER
->__count
));
295 write_unlock_irq(&tasklist_lock
);
296 switch_uid(INIT_USER
);
299 void __set_special_pids(pid_t session
, pid_t pgrp
)
301 struct task_struct
*curr
= current
->group_leader
;
303 if (process_session(curr
) != session
) {
304 detach_pid(curr
, PIDTYPE_SID
);
305 set_signal_session(curr
->signal
, session
);
306 attach_pid(curr
, PIDTYPE_SID
, session
);
308 if (process_group(curr
) != pgrp
) {
309 detach_pid(curr
, PIDTYPE_PGID
);
310 curr
->signal
->pgrp
= pgrp
;
311 attach_pid(curr
, PIDTYPE_PGID
, pgrp
);
315 static void set_special_pids(pid_t session
, pid_t pgrp
)
317 write_lock_irq(&tasklist_lock
);
318 __set_special_pids(session
, pgrp
);
319 write_unlock_irq(&tasklist_lock
);
323 * Let kernel threads use this to say that they
324 * allow a certain signal (since daemonize() will
325 * have disabled all of them by default).
327 int allow_signal(int sig
)
329 if (!valid_signal(sig
) || sig
< 1)
332 spin_lock_irq(¤t
->sighand
->siglock
);
333 sigdelset(¤t
->blocked
, sig
);
335 /* Kernel threads handle their own signals.
336 Let the signal code know it'll be handled, so
337 that they don't get converted to SIGKILL or
338 just silently dropped */
339 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= (void __user
*)2;
342 spin_unlock_irq(¤t
->sighand
->siglock
);
346 EXPORT_SYMBOL(allow_signal
);
348 int disallow_signal(int sig
)
350 if (!valid_signal(sig
) || sig
< 1)
353 spin_lock_irq(¤t
->sighand
->siglock
);
354 sigaddset(¤t
->blocked
, sig
);
356 spin_unlock_irq(¤t
->sighand
->siglock
);
360 EXPORT_SYMBOL(disallow_signal
);
363 * Put all the gunge required to become a kernel thread without
364 * attached user resources in one place where it belongs.
367 void daemonize(const char *name
, ...)
370 struct fs_struct
*fs
;
373 va_start(args
, name
);
374 vsnprintf(current
->comm
, sizeof(current
->comm
), name
, args
);
378 * If we were started as result of loading a module, close all of the
379 * user space pages. We don't need them, and if we didn't close them
380 * they would be locked into memory.
384 set_special_pids(1, 1);
385 proc_clear_tty(current
);
387 /* Block and flush all signals */
388 sigfillset(&blocked
);
389 sigprocmask(SIG_BLOCK
, &blocked
, NULL
);
390 flush_signals(current
);
392 /* Become as one with the init task */
394 exit_fs(current
); /* current->fs->count--; */
397 atomic_inc(&fs
->count
);
399 exit_task_namespaces(current
);
400 current
->nsproxy
= init_task
.nsproxy
;
401 get_task_namespaces(current
);
404 current
->files
= init_task
.files
;
405 atomic_inc(¤t
->files
->count
);
410 EXPORT_SYMBOL(daemonize
);
412 static void close_files(struct files_struct
* files
)
420 * It is safe to dereference the fd table without RCU or
421 * ->file_lock because this is the last reference to the
424 fdt
= files_fdtable(files
);
428 if (i
>= fdt
->max_fds
)
430 set
= fdt
->open_fds
->fds_bits
[j
++];
433 struct file
* file
= xchg(&fdt
->fd
[i
], NULL
);
435 filp_close(file
, files
);
443 struct files_struct
*get_files_struct(struct task_struct
*task
)
445 struct files_struct
*files
;
450 atomic_inc(&files
->count
);
456 void fastcall
put_files_struct(struct files_struct
*files
)
460 if (atomic_dec_and_test(&files
->count
)) {
463 * Free the fd and fdset arrays if we expanded them.
464 * If the fdtable was embedded, pass files for freeing
465 * at the end of the RCU grace period. Otherwise,
466 * you can free files immediately.
468 fdt
= files_fdtable(files
);
469 if (fdt
!= &files
->fdtab
)
470 kmem_cache_free(files_cachep
, files
);
475 EXPORT_SYMBOL(put_files_struct
);
477 void reset_files_struct(struct task_struct
*tsk
, struct files_struct
*files
)
479 struct files_struct
*old
;
485 put_files_struct(old
);
487 EXPORT_SYMBOL(reset_files_struct
);
489 static inline void __exit_files(struct task_struct
*tsk
)
491 struct files_struct
* files
= tsk
->files
;
497 put_files_struct(files
);
501 void exit_files(struct task_struct
*tsk
)
506 static inline void __put_fs_struct(struct fs_struct
*fs
)
508 /* No need to hold fs->lock if we are killing it */
509 if (atomic_dec_and_test(&fs
->count
)) {
516 mntput(fs
->altrootmnt
);
518 kmem_cache_free(fs_cachep
, fs
);
522 void put_fs_struct(struct fs_struct
*fs
)
527 static inline void __exit_fs(struct task_struct
*tsk
)
529 struct fs_struct
* fs
= tsk
->fs
;
539 void exit_fs(struct task_struct
*tsk
)
544 EXPORT_SYMBOL_GPL(exit_fs
);
547 * Turn us into a lazy TLB process if we
550 static void exit_mm(struct task_struct
* tsk
)
552 struct mm_struct
*mm
= tsk
->mm
;
558 * Serialize with any possible pending coredump.
559 * We must hold mmap_sem around checking core_waiters
560 * and clearing tsk->mm. The core-inducing thread
561 * will increment core_waiters for each thread in the
562 * group with ->mm != NULL.
564 down_read(&mm
->mmap_sem
);
565 if (mm
->core_waiters
) {
566 up_read(&mm
->mmap_sem
);
567 down_write(&mm
->mmap_sem
);
568 if (!--mm
->core_waiters
)
569 complete(mm
->core_startup_done
);
570 up_write(&mm
->mmap_sem
);
572 wait_for_completion(&mm
->core_done
);
573 down_read(&mm
->mmap_sem
);
575 atomic_inc(&mm
->mm_count
);
576 BUG_ON(mm
!= tsk
->active_mm
);
577 /* more a memory barrier than a real lock */
580 up_read(&mm
->mmap_sem
);
581 enter_lazy_tlb(mm
, current
);
587 choose_new_parent(struct task_struct
*p
, struct task_struct
*reaper
)
590 * Make sure we're not reparenting to ourselves and that
591 * the parent is not a zombie.
593 BUG_ON(p
== reaper
|| reaper
->exit_state
);
594 p
->real_parent
= reaper
;
598 reparent_thread(struct task_struct
*p
, struct task_struct
*father
, int traced
)
600 if (p
->pdeath_signal
)
601 /* We already hold the tasklist_lock here. */
602 group_send_sig_info(p
->pdeath_signal
, SEND_SIG_NOINFO
, p
);
604 /* Move the child from its dying parent to the new one. */
605 if (unlikely(traced
)) {
606 /* Preserve ptrace links if someone else is tracing this child. */
607 list_del_init(&p
->ptrace_list
);
608 if (p
->parent
!= p
->real_parent
)
609 list_add(&p
->ptrace_list
, &p
->real_parent
->ptrace_children
);
611 /* If this child is being traced, then we're the one tracing it
612 * anyway, so let go of it.
616 p
->parent
= p
->real_parent
;
619 if (p
->state
== TASK_TRACED
) {
621 * If it was at a trace stop, turn it into
622 * a normal stop since it's no longer being
629 /* If this is a threaded reparent there is no need to
630 * notify anyone anything has happened.
632 if (p
->real_parent
->group_leader
== father
->group_leader
)
635 /* We don't want people slaying init. */
636 if (p
->exit_signal
!= -1)
637 p
->exit_signal
= SIGCHLD
;
639 /* If we'd notified the old parent about this child's death,
640 * also notify the new parent.
642 if (!traced
&& p
->exit_state
== EXIT_ZOMBIE
&&
643 p
->exit_signal
!= -1 && thread_group_empty(p
))
644 do_notify_parent(p
, p
->exit_signal
);
647 * process group orphan check
648 * Case ii: Our child is in a different pgrp
649 * than we are, and it was the only connection
650 * outside, so the child pgrp is now orphaned.
652 if ((process_group(p
) != process_group(father
)) &&
653 (process_session(p
) == process_session(father
))) {
654 int pgrp
= process_group(p
);
656 if (will_become_orphaned_pgrp(pgrp
, NULL
) &&
657 has_stopped_jobs(pgrp
)) {
658 __kill_pg_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
659 __kill_pg_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
665 * When we die, we re-parent all our children.
666 * Try to give them to another thread in our thread
667 * group, and if no such member exists, give it to
668 * the child reaper process (ie "init") in our pid
672 forget_original_parent(struct task_struct
*father
, struct list_head
*to_release
)
674 struct task_struct
*p
, *reaper
= father
;
675 struct list_head
*_p
, *_n
;
678 reaper
= next_thread(reaper
);
679 if (reaper
== father
) {
680 reaper
= child_reaper(father
);
683 } while (reaper
->exit_state
);
686 * There are only two places where our children can be:
688 * - in our child list
689 * - in our ptraced child list
691 * Search them and reparent children.
693 list_for_each_safe(_p
, _n
, &father
->children
) {
695 p
= list_entry(_p
, struct task_struct
, sibling
);
699 /* if father isn't the real parent, then ptrace must be enabled */
700 BUG_ON(father
!= p
->real_parent
&& !ptrace
);
702 if (father
== p
->real_parent
) {
703 /* reparent with a reaper, real father it's us */
704 choose_new_parent(p
, reaper
);
705 reparent_thread(p
, father
, 0);
707 /* reparent ptraced task to its real parent */
709 if (p
->exit_state
== EXIT_ZOMBIE
&& p
->exit_signal
!= -1 &&
710 thread_group_empty(p
))
711 do_notify_parent(p
, p
->exit_signal
);
715 * if the ptraced child is a zombie with exit_signal == -1
716 * we must collect it before we exit, or it will remain
717 * zombie forever since we prevented it from self-reap itself
718 * while it was being traced by us, to be able to see it in wait4.
720 if (unlikely(ptrace
&& p
->exit_state
== EXIT_ZOMBIE
&& p
->exit_signal
== -1))
721 list_add(&p
->ptrace_list
, to_release
);
723 list_for_each_safe(_p
, _n
, &father
->ptrace_children
) {
724 p
= list_entry(_p
, struct task_struct
, ptrace_list
);
725 choose_new_parent(p
, reaper
);
726 reparent_thread(p
, father
, 1);
731 * Send signals to all our closest relatives so that they know
732 * to properly mourn us..
734 static void exit_notify(struct task_struct
*tsk
)
737 struct task_struct
*t
;
738 struct list_head ptrace_dead
, *_p
, *_n
;
740 if (signal_pending(tsk
) && !(tsk
->signal
->flags
& SIGNAL_GROUP_EXIT
)
741 && !thread_group_empty(tsk
)) {
743 * This occurs when there was a race between our exit
744 * syscall and a group signal choosing us as the one to
745 * wake up. It could be that we are the only thread
746 * alerted to check for pending signals, but another thread
747 * should be woken now to take the signal since we will not.
748 * Now we'll wake all the threads in the group just to make
749 * sure someone gets all the pending signals.
751 read_lock(&tasklist_lock
);
752 spin_lock_irq(&tsk
->sighand
->siglock
);
753 for (t
= next_thread(tsk
); t
!= tsk
; t
= next_thread(t
))
754 if (!signal_pending(t
) && !(t
->flags
& PF_EXITING
)) {
755 recalc_sigpending_tsk(t
);
756 if (signal_pending(t
))
757 signal_wake_up(t
, 0);
759 spin_unlock_irq(&tsk
->sighand
->siglock
);
760 read_unlock(&tasklist_lock
);
763 write_lock_irq(&tasklist_lock
);
766 * This does two things:
768 * A. Make init inherit all the child processes
769 * B. Check to see if any process groups have become orphaned
770 * as a result of our exiting, and if they have any stopped
771 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
774 INIT_LIST_HEAD(&ptrace_dead
);
775 forget_original_parent(tsk
, &ptrace_dead
);
776 BUG_ON(!list_empty(&tsk
->children
));
777 BUG_ON(!list_empty(&tsk
->ptrace_children
));
780 * Check to see if any process groups have become orphaned
781 * as a result of our exiting, and if they have any stopped
782 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
784 * Case i: Our father is in a different pgrp than we are
785 * and we were the only connection outside, so our pgrp
786 * is about to become orphaned.
789 t
= tsk
->real_parent
;
791 if ((process_group(t
) != process_group(tsk
)) &&
792 (process_session(t
) == process_session(tsk
)) &&
793 will_become_orphaned_pgrp(process_group(tsk
), tsk
) &&
794 has_stopped_jobs(process_group(tsk
))) {
795 __kill_pg_info(SIGHUP
, SEND_SIG_PRIV
, process_group(tsk
));
796 __kill_pg_info(SIGCONT
, SEND_SIG_PRIV
, process_group(tsk
));
799 /* Let father know we died
801 * Thread signals are configurable, but you aren't going to use
802 * that to send signals to arbitary processes.
803 * That stops right now.
805 * If the parent exec id doesn't match the exec id we saved
806 * when we started then we know the parent has changed security
809 * If our self_exec id doesn't match our parent_exec_id then
810 * we have changed execution domain as these two values started
811 * the same after a fork.
815 if (tsk
->exit_signal
!= SIGCHLD
&& tsk
->exit_signal
!= -1 &&
816 ( tsk
->parent_exec_id
!= t
->self_exec_id
||
817 tsk
->self_exec_id
!= tsk
->parent_exec_id
)
818 && !capable(CAP_KILL
))
819 tsk
->exit_signal
= SIGCHLD
;
822 /* If something other than our normal parent is ptracing us, then
823 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
824 * only has special meaning to our real parent.
826 if (tsk
->exit_signal
!= -1 && thread_group_empty(tsk
)) {
827 int signal
= tsk
->parent
== tsk
->real_parent
? tsk
->exit_signal
: SIGCHLD
;
828 do_notify_parent(tsk
, signal
);
829 } else if (tsk
->ptrace
) {
830 do_notify_parent(tsk
, SIGCHLD
);
834 if (tsk
->exit_signal
== -1 &&
835 (likely(tsk
->ptrace
== 0) ||
836 unlikely(tsk
->parent
->signal
->flags
& SIGNAL_GROUP_EXIT
)))
838 tsk
->exit_state
= state
;
840 write_unlock_irq(&tasklist_lock
);
842 list_for_each_safe(_p
, _n
, &ptrace_dead
) {
844 t
= list_entry(_p
, struct task_struct
, ptrace_list
);
848 /* If the process is dead, release it - nobody will wait for it */
849 if (state
== EXIT_DEAD
)
853 fastcall NORET_TYPE
void do_exit(long code
)
855 struct task_struct
*tsk
= current
;
858 profile_task_exit(tsk
);
860 WARN_ON(atomic_read(&tsk
->fs_excl
));
862 if (unlikely(in_interrupt()))
863 panic("Aiee, killing interrupt handler!");
864 if (unlikely(!tsk
->pid
))
865 panic("Attempted to kill the idle task!");
866 if (unlikely(tsk
== child_reaper(tsk
))) {
867 if (tsk
->nsproxy
->pid_ns
!= &init_pid_ns
)
868 tsk
->nsproxy
->pid_ns
->child_reaper
= init_pid_ns
.child_reaper
;
870 panic("Attempted to kill init!");
874 if (unlikely(current
->ptrace
& PT_TRACE_EXIT
)) {
875 current
->ptrace_message
= code
;
876 ptrace_notify((PTRACE_EVENT_EXIT
<< 8) | SIGTRAP
);
880 * We're taking recursive faults here in do_exit. Safest is to just
881 * leave this task alone and wait for reboot.
883 if (unlikely(tsk
->flags
& PF_EXITING
)) {
885 "Fixing recursive fault but reboot is needed!\n");
887 * We can do this unlocked here. The futex code uses
888 * this flag just to verify whether the pi state
889 * cleanup has been done or not. In the worst case it
890 * loops once more. We pretend that the cleanup was
891 * done as there is no way to return. Either the
892 * OWNER_DIED bit is set by now or we push the blocked
893 * task into the wait for ever nirwana as well.
895 tsk
->flags
|= PF_EXITPIDONE
;
898 set_current_state(TASK_UNINTERRUPTIBLE
);
903 * tsk->flags are checked in the futex code to protect against
904 * an exiting task cleaning up the robust pi futexes.
906 spin_lock_irq(&tsk
->pi_lock
);
907 tsk
->flags
|= PF_EXITING
;
908 spin_unlock_irq(&tsk
->pi_lock
);
910 if (unlikely(in_atomic()))
911 printk(KERN_INFO
"note: %s[%d] exited with preempt_count %d\n",
912 current
->comm
, current
->pid
,
915 acct_update_integrals(tsk
);
917 update_hiwater_rss(tsk
->mm
);
918 update_hiwater_vm(tsk
->mm
);
920 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
922 hrtimer_cancel(&tsk
->signal
->real_timer
);
923 exit_itimers(tsk
->signal
);
925 acct_collect(code
, group_dead
);
926 if (unlikely(tsk
->robust_list
))
927 exit_robust_list(tsk
);
928 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
929 if (unlikely(tsk
->compat_robust_list
))
930 compat_exit_robust_list(tsk
);
932 if (unlikely(tsk
->audit_context
))
935 taskstats_exit(tsk
, group_dead
);
948 if (group_dead
&& tsk
->signal
->leader
)
949 disassociate_ctty(1);
951 module_put(task_thread_info(tsk
)->exec_domain
->module
);
953 module_put(tsk
->binfmt
->module
);
955 tsk
->exit_code
= code
;
956 proc_exit_connector(tsk
);
957 exit_task_namespaces(tsk
);
960 mpol_free(tsk
->mempolicy
);
961 tsk
->mempolicy
= NULL
;
964 * This must happen late, after the PID is not
967 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
968 exit_pi_state_list(tsk
);
969 if (unlikely(current
->pi_state_cache
))
970 kfree(current
->pi_state_cache
);
972 * Make sure we are holding no locks:
974 debug_check_no_locks_held(tsk
);
976 * We can do this unlocked here. The futex code uses this flag
977 * just to verify whether the pi state cleanup has been done
978 * or not. In the worst case it loops once more.
980 tsk
->flags
|= PF_EXITPIDONE
;
985 if (tsk
->splice_pipe
)
986 __free_pipe_info(tsk
->splice_pipe
);
989 /* causes final put_task_struct in finish_task_switch(). */
990 tsk
->state
= TASK_DEAD
;
994 /* Avoid "noreturn function does return". */
996 cpu_relax(); /* For when BUG is null */
999 EXPORT_SYMBOL_GPL(do_exit
);
1001 NORET_TYPE
void complete_and_exit(struct completion
*comp
, long code
)
1009 EXPORT_SYMBOL(complete_and_exit
);
1011 asmlinkage
long sys_exit(int error_code
)
1013 do_exit((error_code
&0xff)<<8);
1017 * Take down every thread in the group. This is called by fatal signals
1018 * as well as by sys_exit_group (below).
1021 do_group_exit(int exit_code
)
1023 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
1025 if (current
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1026 exit_code
= current
->signal
->group_exit_code
;
1027 else if (!thread_group_empty(current
)) {
1028 struct signal_struct
*const sig
= current
->signal
;
1029 struct sighand_struct
*const sighand
= current
->sighand
;
1030 spin_lock_irq(&sighand
->siglock
);
1031 if (sig
->flags
& SIGNAL_GROUP_EXIT
)
1032 /* Another thread got here before we took the lock. */
1033 exit_code
= sig
->group_exit_code
;
1035 sig
->group_exit_code
= exit_code
;
1036 zap_other_threads(current
);
1038 spin_unlock_irq(&sighand
->siglock
);
1046 * this kills every thread in the thread group. Note that any externally
1047 * wait4()-ing process will get the correct exit code - even if this
1048 * thread is not the thread group leader.
1050 asmlinkage
void sys_exit_group(int error_code
)
1052 do_group_exit((error_code
& 0xff) << 8);
1055 static int eligible_child(pid_t pid
, int options
, struct task_struct
*p
)
1061 if (process_group(p
) != process_group(current
))
1063 } else if (pid
!= -1) {
1064 if (process_group(p
) != -pid
)
1069 * Do not consider detached threads that are
1072 if (p
->exit_signal
== -1 && !p
->ptrace
)
1075 /* Wait for all children (clone and not) if __WALL is set;
1076 * otherwise, wait for clone children *only* if __WCLONE is
1077 * set; otherwise, wait for non-clone children *only*. (Note:
1078 * A "clone" child here is one that reports to its parent
1079 * using a signal other than SIGCHLD.) */
1080 if (((p
->exit_signal
!= SIGCHLD
) ^ ((options
& __WCLONE
) != 0))
1081 && !(options
& __WALL
))
1084 * Do not consider thread group leaders that are
1085 * in a non-empty thread group:
1087 if (delay_group_leader(p
))
1090 if (security_task_wait(p
))
1096 static int wait_noreap_copyout(struct task_struct
*p
, pid_t pid
, uid_t uid
,
1097 int why
, int status
,
1098 struct siginfo __user
*infop
,
1099 struct rusage __user
*rusagep
)
1101 int retval
= rusagep
? getrusage(p
, RUSAGE_BOTH
, rusagep
) : 0;
1105 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1107 retval
= put_user(0, &infop
->si_errno
);
1109 retval
= put_user((short)why
, &infop
->si_code
);
1111 retval
= put_user(pid
, &infop
->si_pid
);
1113 retval
= put_user(uid
, &infop
->si_uid
);
1115 retval
= put_user(status
, &infop
->si_status
);
1122 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1123 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1124 * the lock and this task is uninteresting. If we return nonzero, we have
1125 * released the lock and the system call should return.
1127 static int wait_task_zombie(struct task_struct
*p
, int noreap
,
1128 struct siginfo __user
*infop
,
1129 int __user
*stat_addr
, struct rusage __user
*ru
)
1131 unsigned long state
;
1135 if (unlikely(noreap
)) {
1138 int exit_code
= p
->exit_code
;
1141 if (unlikely(p
->exit_state
!= EXIT_ZOMBIE
))
1143 if (unlikely(p
->exit_signal
== -1 && p
->ptrace
== 0))
1146 read_unlock(&tasklist_lock
);
1147 if ((exit_code
& 0x7f) == 0) {
1149 status
= exit_code
>> 8;
1151 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1152 status
= exit_code
& 0x7f;
1154 return wait_noreap_copyout(p
, pid
, uid
, why
,
1159 * Try to move the task's state to DEAD
1160 * only one thread is allowed to do this:
1162 state
= xchg(&p
->exit_state
, EXIT_DEAD
);
1163 if (state
!= EXIT_ZOMBIE
) {
1164 BUG_ON(state
!= EXIT_DEAD
);
1167 if (unlikely(p
->exit_signal
== -1 && p
->ptrace
== 0)) {
1169 * This can only happen in a race with a ptraced thread
1170 * dying on another processor.
1175 if (likely(p
->real_parent
== p
->parent
) && likely(p
->signal
)) {
1176 struct signal_struct
*psig
;
1177 struct signal_struct
*sig
;
1180 * The resource counters for the group leader are in its
1181 * own task_struct. Those for dead threads in the group
1182 * are in its signal_struct, as are those for the child
1183 * processes it has previously reaped. All these
1184 * accumulate in the parent's signal_struct c* fields.
1186 * We don't bother to take a lock here to protect these
1187 * p->signal fields, because they are only touched by
1188 * __exit_signal, which runs with tasklist_lock
1189 * write-locked anyway, and so is excluded here. We do
1190 * need to protect the access to p->parent->signal fields,
1191 * as other threads in the parent group can be right
1192 * here reaping other children at the same time.
1194 spin_lock_irq(&p
->parent
->sighand
->siglock
);
1195 psig
= p
->parent
->signal
;
1198 cputime_add(psig
->cutime
,
1199 cputime_add(p
->utime
,
1200 cputime_add(sig
->utime
,
1203 cputime_add(psig
->cstime
,
1204 cputime_add(p
->stime
,
1205 cputime_add(sig
->stime
,
1208 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1210 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1212 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1214 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1215 spin_unlock_irq(&p
->parent
->sighand
->siglock
);
1219 * Now we are sure this task is interesting, and no other
1220 * thread can reap it because we set its state to EXIT_DEAD.
1222 read_unlock(&tasklist_lock
);
1224 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1225 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1226 ? p
->signal
->group_exit_code
: p
->exit_code
;
1227 if (!retval
&& stat_addr
)
1228 retval
= put_user(status
, stat_addr
);
1229 if (!retval
&& infop
)
1230 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1231 if (!retval
&& infop
)
1232 retval
= put_user(0, &infop
->si_errno
);
1233 if (!retval
&& infop
) {
1236 if ((status
& 0x7f) == 0) {
1240 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1243 retval
= put_user((short)why
, &infop
->si_code
);
1245 retval
= put_user(status
, &infop
->si_status
);
1247 if (!retval
&& infop
)
1248 retval
= put_user(p
->pid
, &infop
->si_pid
);
1249 if (!retval
&& infop
)
1250 retval
= put_user(p
->uid
, &infop
->si_uid
);
1252 // TODO: is this safe?
1253 p
->exit_state
= EXIT_ZOMBIE
;
1257 if (p
->real_parent
!= p
->parent
) {
1258 write_lock_irq(&tasklist_lock
);
1259 /* Double-check with lock held. */
1260 if (p
->real_parent
!= p
->parent
) {
1262 // TODO: is this safe?
1263 p
->exit_state
= EXIT_ZOMBIE
;
1265 * If this is not a detached task, notify the parent.
1266 * If it's still not detached after that, don't release
1269 if (p
->exit_signal
!= -1) {
1270 do_notify_parent(p
, p
->exit_signal
);
1271 if (p
->exit_signal
!= -1)
1275 write_unlock_irq(&tasklist_lock
);
1284 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1285 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1286 * the lock and this task is uninteresting. If we return nonzero, we have
1287 * released the lock and the system call should return.
1289 static int wait_task_stopped(struct task_struct
*p
, int delayed_group_leader
,
1290 int noreap
, struct siginfo __user
*infop
,
1291 int __user
*stat_addr
, struct rusage __user
*ru
)
1293 int retval
, exit_code
;
1297 if (delayed_group_leader
&& !(p
->ptrace
& PT_PTRACED
) &&
1298 p
->signal
&& p
->signal
->group_stop_count
> 0)
1300 * A group stop is in progress and this is the group leader.
1301 * We won't report until all threads have stopped.
1306 * Now we are pretty sure this task is interesting.
1307 * Make sure it doesn't get reaped out from under us while we
1308 * give up the lock and then examine it below. We don't want to
1309 * keep holding onto the tasklist_lock while we call getrusage and
1310 * possibly take page faults for user memory.
1313 read_unlock(&tasklist_lock
);
1315 if (unlikely(noreap
)) {
1318 int why
= (p
->ptrace
& PT_PTRACED
) ? CLD_TRAPPED
: CLD_STOPPED
;
1320 exit_code
= p
->exit_code
;
1321 if (unlikely(!exit_code
) ||
1322 unlikely(p
->state
& TASK_TRACED
))
1324 return wait_noreap_copyout(p
, pid
, uid
,
1325 why
, (exit_code
<< 8) | 0x7f,
1329 write_lock_irq(&tasklist_lock
);
1332 * This uses xchg to be atomic with the thread resuming and setting
1333 * it. It must also be done with the write lock held to prevent a
1334 * race with the EXIT_ZOMBIE case.
1336 exit_code
= xchg(&p
->exit_code
, 0);
1337 if (unlikely(p
->exit_state
)) {
1339 * The task resumed and then died. Let the next iteration
1340 * catch it in EXIT_ZOMBIE. Note that exit_code might
1341 * already be zero here if it resumed and did _exit(0).
1342 * The task itself is dead and won't touch exit_code again;
1343 * other processors in this function are locked out.
1345 p
->exit_code
= exit_code
;
1348 if (unlikely(exit_code
== 0)) {
1350 * Another thread in this function got to it first, or it
1351 * resumed, or it resumed and then died.
1353 write_unlock_irq(&tasklist_lock
);
1357 * We are returning to the wait loop without having successfully
1358 * removed the process and having released the lock. We cannot
1359 * continue, since the "p" task pointer is potentially stale.
1361 * Return -EAGAIN, and do_wait() will restart the loop from the
1362 * beginning. Do _not_ re-acquire the lock.
1367 /* move to end of parent's list to avoid starvation */
1371 write_unlock_irq(&tasklist_lock
);
1373 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1374 if (!retval
&& stat_addr
)
1375 retval
= put_user((exit_code
<< 8) | 0x7f, stat_addr
);
1376 if (!retval
&& infop
)
1377 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1378 if (!retval
&& infop
)
1379 retval
= put_user(0, &infop
->si_errno
);
1380 if (!retval
&& infop
)
1381 retval
= put_user((short)((p
->ptrace
& PT_PTRACED
)
1382 ? CLD_TRAPPED
: CLD_STOPPED
),
1384 if (!retval
&& infop
)
1385 retval
= put_user(exit_code
, &infop
->si_status
);
1386 if (!retval
&& infop
)
1387 retval
= put_user(p
->pid
, &infop
->si_pid
);
1388 if (!retval
&& infop
)
1389 retval
= put_user(p
->uid
, &infop
->si_uid
);
1399 * Handle do_wait work for one task in a live, non-stopped state.
1400 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1401 * the lock and this task is uninteresting. If we return nonzero, we have
1402 * released the lock and the system call should return.
1404 static int wait_task_continued(struct task_struct
*p
, int noreap
,
1405 struct siginfo __user
*infop
,
1406 int __user
*stat_addr
, struct rusage __user
*ru
)
1412 if (unlikely(!p
->signal
))
1415 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1418 spin_lock_irq(&p
->sighand
->siglock
);
1419 /* Re-check with the lock held. */
1420 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1421 spin_unlock_irq(&p
->sighand
->siglock
);
1425 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1426 spin_unlock_irq(&p
->sighand
->siglock
);
1431 read_unlock(&tasklist_lock
);
1434 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1436 if (!retval
&& stat_addr
)
1437 retval
= put_user(0xffff, stat_addr
);
1441 retval
= wait_noreap_copyout(p
, pid
, uid
,
1442 CLD_CONTINUED
, SIGCONT
,
1444 BUG_ON(retval
== 0);
1451 static inline int my_ptrace_child(struct task_struct
*p
)
1453 if (!(p
->ptrace
& PT_PTRACED
))
1455 if (!(p
->ptrace
& PT_ATTACHED
))
1458 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1459 * we are the attacher. If we are the real parent, this is a race
1460 * inside ptrace_attach. It is waiting for the tasklist_lock,
1461 * which we have to switch the parent links, but has already set
1462 * the flags in p->ptrace.
1464 return (p
->parent
!= p
->real_parent
);
1467 static long do_wait(pid_t pid
, int options
, struct siginfo __user
*infop
,
1468 int __user
*stat_addr
, struct rusage __user
*ru
)
1470 DECLARE_WAITQUEUE(wait
, current
);
1471 struct task_struct
*tsk
;
1474 add_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1477 * We will set this flag if we see any child that might later
1478 * match our criteria, even if we are not able to reap it yet.
1481 current
->state
= TASK_INTERRUPTIBLE
;
1482 read_lock(&tasklist_lock
);
1485 struct task_struct
*p
;
1486 struct list_head
*_p
;
1489 list_for_each(_p
,&tsk
->children
) {
1490 p
= list_entry(_p
, struct task_struct
, sibling
);
1492 ret
= eligible_child(pid
, options
, p
);
1499 * When we hit the race with PTRACE_ATTACH,
1500 * we will not report this child. But the
1501 * race means it has not yet been moved to
1502 * our ptrace_children list, so we need to
1503 * set the flag here to avoid a spurious ECHILD
1504 * when the race happens with the only child.
1507 if (!my_ptrace_child(p
))
1512 * It's stopped now, so it might later
1513 * continue, exit, or stop again.
1516 if (!(options
& WUNTRACED
) &&
1517 !my_ptrace_child(p
))
1519 retval
= wait_task_stopped(p
, ret
== 2,
1520 (options
& WNOWAIT
),
1523 if (retval
== -EAGAIN
)
1525 if (retval
!= 0) /* He released the lock. */
1530 if (p
->exit_state
== EXIT_DEAD
)
1532 // case EXIT_ZOMBIE:
1533 if (p
->exit_state
== EXIT_ZOMBIE
) {
1535 * Eligible but we cannot release
1539 goto check_continued
;
1540 if (!likely(options
& WEXITED
))
1542 retval
= wait_task_zombie(
1543 p
, (options
& WNOWAIT
),
1544 infop
, stat_addr
, ru
);
1545 /* He released the lock. */
1552 * It's running now, so it might later
1553 * exit, stop, or stop and then continue.
1556 if (!unlikely(options
& WCONTINUED
))
1558 retval
= wait_task_continued(
1559 p
, (options
& WNOWAIT
),
1560 infop
, stat_addr
, ru
);
1561 if (retval
!= 0) /* He released the lock. */
1567 list_for_each(_p
, &tsk
->ptrace_children
) {
1568 p
= list_entry(_p
, struct task_struct
,
1570 if (!eligible_child(pid
, options
, p
))
1576 if (options
& __WNOTHREAD
)
1578 tsk
= next_thread(tsk
);
1579 BUG_ON(tsk
->signal
!= current
->signal
);
1580 } while (tsk
!= current
);
1582 read_unlock(&tasklist_lock
);
1585 if (options
& WNOHANG
)
1587 retval
= -ERESTARTSYS
;
1588 if (signal_pending(current
))
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
);