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/signalfd.h>
28 #include <linux/mount.h>
29 #include <linux/proc_fs.h>
30 #include <linux/kthread.h>
31 #include <linux/mempolicy.h>
32 #include <linux/taskstats_kern.h>
33 #include <linux/delayacct.h>
34 #include <linux/cpuset.h>
35 #include <linux/syscalls.h>
36 #include <linux/signal.h>
37 #include <linux/posix-timers.h>
38 #include <linux/cn_proc.h>
39 #include <linux/mutex.h>
40 #include <linux/futex.h>
41 #include <linux/compat.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
48 #include <asm/uaccess.h>
49 #include <asm/unistd.h>
50 #include <asm/pgtable.h>
51 #include <asm/mmu_context.h>
53 extern void sem_exit (void);
55 static void exit_mm(struct task_struct
* tsk
);
57 static void __unhash_process(struct task_struct
*p
)
60 detach_pid(p
, PIDTYPE_PID
);
61 if (thread_group_leader(p
)) {
62 detach_pid(p
, PIDTYPE_PGID
);
63 detach_pid(p
, PIDTYPE_SID
);
65 list_del_rcu(&p
->tasks
);
66 __get_cpu_var(process_counts
)--;
68 list_del_rcu(&p
->thread_group
);
73 * This function expects the tasklist_lock write-locked.
75 static void __exit_signal(struct task_struct
*tsk
)
77 struct signal_struct
*sig
= tsk
->signal
;
78 struct sighand_struct
*sighand
;
81 BUG_ON(!atomic_read(&sig
->count
));
84 sighand
= rcu_dereference(tsk
->sighand
);
85 spin_lock(&sighand
->siglock
);
88 * Notify that this sighand has been detached. This must
89 * be called with the tsk->sighand lock held. Also, this
90 * access tsk->sighand internally, so it must be called
91 * before tsk->sighand is reset.
93 signalfd_detach_locked(tsk
);
95 posix_cpu_timers_exit(tsk
);
96 if (atomic_dec_and_test(&sig
->count
))
97 posix_cpu_timers_exit_group(tsk
);
100 * If there is any task waiting for the group exit
103 if (sig
->group_exit_task
&& atomic_read(&sig
->count
) == sig
->notify_count
) {
104 wake_up_process(sig
->group_exit_task
);
105 sig
->group_exit_task
= NULL
;
107 if (tsk
== sig
->curr_target
)
108 sig
->curr_target
= next_thread(tsk
);
110 * Accumulate here the counters for all threads but the
111 * group leader as they die, so they can be added into
112 * the process-wide totals when those are taken.
113 * The group leader stays around as a zombie as long
114 * as there are other threads. When it gets reaped,
115 * the exit.c code will add its counts into these totals.
116 * We won't ever get here for the group leader, since it
117 * will have been the last reference on the signal_struct.
119 sig
->utime
= cputime_add(sig
->utime
, tsk
->utime
);
120 sig
->stime
= cputime_add(sig
->stime
, tsk
->stime
);
121 sig
->min_flt
+= tsk
->min_flt
;
122 sig
->maj_flt
+= tsk
->maj_flt
;
123 sig
->nvcsw
+= tsk
->nvcsw
;
124 sig
->nivcsw
+= tsk
->nivcsw
;
125 sig
->inblock
+= task_io_get_inblock(tsk
);
126 sig
->oublock
+= task_io_get_oublock(tsk
);
127 sig
->sum_sched_runtime
+= tsk
->se
.sum_exec_runtime
;
128 sig
= NULL
; /* Marker for below. */
131 __unhash_process(tsk
);
135 spin_unlock(&sighand
->siglock
);
138 __cleanup_sighand(sighand
);
139 clear_tsk_thread_flag(tsk
,TIF_SIGPENDING
);
140 flush_sigqueue(&tsk
->pending
);
142 flush_sigqueue(&sig
->shared_pending
);
143 taskstats_tgid_free(sig
);
144 __cleanup_signal(sig
);
148 static void delayed_put_task_struct(struct rcu_head
*rhp
)
150 put_task_struct(container_of(rhp
, struct task_struct
, rcu
));
153 void release_task(struct task_struct
* p
)
155 struct task_struct
*leader
;
158 atomic_dec(&p
->user
->processes
);
159 write_lock_irq(&tasklist_lock
);
161 BUG_ON(!list_empty(&p
->ptrace_list
) || !list_empty(&p
->ptrace_children
));
165 * If we are the last non-leader member of the thread
166 * group, and the leader is zombie, then notify the
167 * group leader's parent process. (if it wants notification.)
170 leader
= p
->group_leader
;
171 if (leader
!= p
&& thread_group_empty(leader
) && leader
->exit_state
== EXIT_ZOMBIE
) {
172 BUG_ON(leader
->exit_signal
== -1);
173 do_notify_parent(leader
, leader
->exit_signal
);
175 * If we were the last child thread and the leader has
176 * exited already, and the leader's parent ignores SIGCHLD,
177 * then we are the one who should release the leader.
179 * do_notify_parent() will have marked it self-reaping in
182 zap_leader
= (leader
->exit_signal
== -1);
185 write_unlock_irq(&tasklist_lock
);
188 call_rcu(&p
->rcu
, delayed_put_task_struct
);
191 if (unlikely(zap_leader
))
196 * This checks not only the pgrp, but falls back on the pid if no
197 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
200 * The caller must hold rcu lock or the tasklist lock.
202 struct pid
*session_of_pgrp(struct pid
*pgrp
)
204 struct task_struct
*p
;
205 struct pid
*sid
= NULL
;
207 p
= pid_task(pgrp
, PIDTYPE_PGID
);
209 p
= pid_task(pgrp
, PIDTYPE_PID
);
211 sid
= task_session(p
);
217 * Determine if a process group is "orphaned", according to the POSIX
218 * definition in 2.2.2.52. Orphaned process groups are not to be affected
219 * by terminal-generated stop signals. Newly orphaned process groups are
220 * to receive a SIGHUP and a SIGCONT.
222 * "I ask you, have you ever known what it is to be an orphan?"
224 static int will_become_orphaned_pgrp(struct pid
*pgrp
, struct task_struct
*ignored_task
)
226 struct task_struct
*p
;
229 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
230 if (p
== ignored_task
232 || is_init(p
->real_parent
))
234 if (task_pgrp(p
->real_parent
) != pgrp
&&
235 task_session(p
->real_parent
) == task_session(p
)) {
239 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
240 return ret
; /* (sighing) "Often!" */
243 int is_current_pgrp_orphaned(void)
247 read_lock(&tasklist_lock
);
248 retval
= will_become_orphaned_pgrp(task_pgrp(current
), NULL
);
249 read_unlock(&tasklist_lock
);
254 static int has_stopped_jobs(struct pid
*pgrp
)
257 struct task_struct
*p
;
259 do_each_pid_task(pgrp
, PIDTYPE_PGID
, p
) {
260 if (p
->state
!= TASK_STOPPED
)
264 } while_each_pid_task(pgrp
, PIDTYPE_PGID
, p
);
269 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
271 * If a kernel thread is launched as a result of a system call, or if
272 * it ever exits, it should generally reparent itself to kthreadd so it
273 * isn't in the way of other processes and is correctly cleaned up on exit.
275 * The various task state such as scheduling policy and priority may have
276 * been inherited from a user process, so we reset them to sane values here.
278 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
280 static void reparent_to_kthreadd(void)
282 write_lock_irq(&tasklist_lock
);
284 ptrace_unlink(current
);
285 /* Reparent to init */
286 remove_parent(current
);
287 current
->real_parent
= current
->parent
= kthreadd_task
;
290 /* Set the exit signal to SIGCHLD so we signal init on exit */
291 current
->exit_signal
= SIGCHLD
;
293 if (task_nice(current
) < 0)
294 set_user_nice(current
, 0);
298 security_task_reparent_to_init(current
);
299 memcpy(current
->signal
->rlim
, init_task
.signal
->rlim
,
300 sizeof(current
->signal
->rlim
));
301 atomic_inc(&(INIT_USER
->__count
));
302 write_unlock_irq(&tasklist_lock
);
303 switch_uid(INIT_USER
);
306 void __set_special_pids(pid_t session
, pid_t pgrp
)
308 struct task_struct
*curr
= current
->group_leader
;
310 if (process_session(curr
) != session
) {
311 detach_pid(curr
, PIDTYPE_SID
);
312 set_signal_session(curr
->signal
, session
);
313 attach_pid(curr
, PIDTYPE_SID
, find_pid(session
));
315 if (process_group(curr
) != pgrp
) {
316 detach_pid(curr
, PIDTYPE_PGID
);
317 curr
->signal
->pgrp
= pgrp
;
318 attach_pid(curr
, PIDTYPE_PGID
, find_pid(pgrp
));
322 static void set_special_pids(pid_t session
, pid_t pgrp
)
324 write_lock_irq(&tasklist_lock
);
325 __set_special_pids(session
, pgrp
);
326 write_unlock_irq(&tasklist_lock
);
330 * Let kernel threads use this to say that they
331 * allow a certain signal (since daemonize() will
332 * have disabled all of them by default).
334 int allow_signal(int sig
)
336 if (!valid_signal(sig
) || sig
< 1)
339 spin_lock_irq(¤t
->sighand
->siglock
);
340 sigdelset(¤t
->blocked
, sig
);
342 /* Kernel threads handle their own signals.
343 Let the signal code know it'll be handled, so
344 that they don't get converted to SIGKILL or
345 just silently dropped */
346 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= (void __user
*)2;
349 spin_unlock_irq(¤t
->sighand
->siglock
);
353 EXPORT_SYMBOL(allow_signal
);
355 int disallow_signal(int sig
)
357 if (!valid_signal(sig
) || sig
< 1)
360 spin_lock_irq(¤t
->sighand
->siglock
);
361 current
->sighand
->action
[(sig
)-1].sa
.sa_handler
= SIG_IGN
;
363 spin_unlock_irq(¤t
->sighand
->siglock
);
367 EXPORT_SYMBOL(disallow_signal
);
370 * Put all the gunge required to become a kernel thread without
371 * attached user resources in one place where it belongs.
374 void daemonize(const char *name
, ...)
377 struct fs_struct
*fs
;
380 va_start(args
, name
);
381 vsnprintf(current
->comm
, sizeof(current
->comm
), name
, args
);
385 * If we were started as result of loading a module, close all of the
386 * user space pages. We don't need them, and if we didn't close them
387 * they would be locked into memory.
391 set_special_pids(1, 1);
392 proc_clear_tty(current
);
394 /* Block and flush all signals */
395 sigfillset(&blocked
);
396 sigprocmask(SIG_BLOCK
, &blocked
, NULL
);
397 flush_signals(current
);
399 /* Become as one with the init task */
401 exit_fs(current
); /* current->fs->count--; */
404 atomic_inc(&fs
->count
);
406 exit_task_namespaces(current
);
407 current
->nsproxy
= init_task
.nsproxy
;
408 get_task_namespaces(current
);
411 current
->files
= init_task
.files
;
412 atomic_inc(¤t
->files
->count
);
414 reparent_to_kthreadd();
417 EXPORT_SYMBOL(daemonize
);
419 static void close_files(struct files_struct
* files
)
427 * It is safe to dereference the fd table without RCU or
428 * ->file_lock because this is the last reference to the
431 fdt
= files_fdtable(files
);
435 if (i
>= fdt
->max_fds
)
437 set
= fdt
->open_fds
->fds_bits
[j
++];
440 struct file
* file
= xchg(&fdt
->fd
[i
], NULL
);
442 filp_close(file
, files
);
452 struct files_struct
*get_files_struct(struct task_struct
*task
)
454 struct files_struct
*files
;
459 atomic_inc(&files
->count
);
465 void fastcall
put_files_struct(struct files_struct
*files
)
469 if (atomic_dec_and_test(&files
->count
)) {
472 * Free the fd and fdset arrays if we expanded them.
473 * If the fdtable was embedded, pass files for freeing
474 * at the end of the RCU grace period. Otherwise,
475 * you can free files immediately.
477 fdt
= files_fdtable(files
);
478 if (fdt
!= &files
->fdtab
)
479 kmem_cache_free(files_cachep
, files
);
484 EXPORT_SYMBOL(put_files_struct
);
486 void reset_files_struct(struct task_struct
*tsk
, struct files_struct
*files
)
488 struct files_struct
*old
;
494 put_files_struct(old
);
496 EXPORT_SYMBOL(reset_files_struct
);
498 static inline void __exit_files(struct task_struct
*tsk
)
500 struct files_struct
* files
= tsk
->files
;
506 put_files_struct(files
);
510 void exit_files(struct task_struct
*tsk
)
515 static inline void __put_fs_struct(struct fs_struct
*fs
)
517 /* No need to hold fs->lock if we are killing it */
518 if (atomic_dec_and_test(&fs
->count
)) {
525 mntput(fs
->altrootmnt
);
527 kmem_cache_free(fs_cachep
, fs
);
531 void put_fs_struct(struct fs_struct
*fs
)
536 static inline void __exit_fs(struct task_struct
*tsk
)
538 struct fs_struct
* fs
= tsk
->fs
;
548 void exit_fs(struct task_struct
*tsk
)
553 EXPORT_SYMBOL_GPL(exit_fs
);
556 * Turn us into a lazy TLB process if we
559 static void exit_mm(struct task_struct
* tsk
)
561 struct mm_struct
*mm
= tsk
->mm
;
567 * Serialize with any possible pending coredump.
568 * We must hold mmap_sem around checking core_waiters
569 * and clearing tsk->mm. The core-inducing thread
570 * will increment core_waiters for each thread in the
571 * group with ->mm != NULL.
573 down_read(&mm
->mmap_sem
);
574 if (mm
->core_waiters
) {
575 up_read(&mm
->mmap_sem
);
576 down_write(&mm
->mmap_sem
);
577 if (!--mm
->core_waiters
)
578 complete(mm
->core_startup_done
);
579 up_write(&mm
->mmap_sem
);
581 wait_for_completion(&mm
->core_done
);
582 down_read(&mm
->mmap_sem
);
584 atomic_inc(&mm
->mm_count
);
585 BUG_ON(mm
!= tsk
->active_mm
);
586 /* more a memory barrier than a real lock */
589 up_read(&mm
->mmap_sem
);
590 enter_lazy_tlb(mm
, current
);
596 choose_new_parent(struct task_struct
*p
, struct task_struct
*reaper
)
599 * Make sure we're not reparenting to ourselves and that
600 * the parent is not a zombie.
602 BUG_ON(p
== reaper
|| reaper
->exit_state
);
603 p
->real_parent
= reaper
;
607 reparent_thread(struct task_struct
*p
, struct task_struct
*father
, int traced
)
609 if (p
->pdeath_signal
)
610 /* We already hold the tasklist_lock here. */
611 group_send_sig_info(p
->pdeath_signal
, SEND_SIG_NOINFO
, p
);
613 /* Move the child from its dying parent to the new one. */
614 if (unlikely(traced
)) {
615 /* Preserve ptrace links if someone else is tracing this child. */
616 list_del_init(&p
->ptrace_list
);
617 if (p
->parent
!= p
->real_parent
)
618 list_add(&p
->ptrace_list
, &p
->real_parent
->ptrace_children
);
620 /* If this child is being traced, then we're the one tracing it
621 * anyway, so let go of it.
625 p
->parent
= p
->real_parent
;
628 if (p
->state
== TASK_TRACED
) {
630 * If it was at a trace stop, turn it into
631 * a normal stop since it's no longer being
638 /* If this is a threaded reparent there is no need to
639 * notify anyone anything has happened.
641 if (p
->real_parent
->group_leader
== father
->group_leader
)
644 /* We don't want people slaying init. */
645 if (p
->exit_signal
!= -1)
646 p
->exit_signal
= SIGCHLD
;
648 /* If we'd notified the old parent about this child's death,
649 * also notify the new parent.
651 if (!traced
&& p
->exit_state
== EXIT_ZOMBIE
&&
652 p
->exit_signal
!= -1 && thread_group_empty(p
))
653 do_notify_parent(p
, p
->exit_signal
);
656 * process group orphan check
657 * Case ii: Our child is in a different pgrp
658 * than we are, and it was the only connection
659 * outside, so the child pgrp is now orphaned.
661 if ((task_pgrp(p
) != task_pgrp(father
)) &&
662 (task_session(p
) == task_session(father
))) {
663 struct pid
*pgrp
= task_pgrp(p
);
665 if (will_become_orphaned_pgrp(pgrp
, NULL
) &&
666 has_stopped_jobs(pgrp
)) {
667 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
668 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
674 * When we die, we re-parent all our children.
675 * Try to give them to another thread in our thread
676 * group, and if no such member exists, give it to
677 * the child reaper process (ie "init") in our pid
681 forget_original_parent(struct task_struct
*father
, struct list_head
*to_release
)
683 struct task_struct
*p
, *reaper
= father
;
684 struct list_head
*_p
, *_n
;
687 reaper
= next_thread(reaper
);
688 if (reaper
== father
) {
689 reaper
= child_reaper(father
);
692 } while (reaper
->exit_state
);
695 * There are only two places where our children can be:
697 * - in our child list
698 * - in our ptraced child list
700 * Search them and reparent children.
702 list_for_each_safe(_p
, _n
, &father
->children
) {
704 p
= list_entry(_p
, struct task_struct
, sibling
);
708 /* if father isn't the real parent, then ptrace must be enabled */
709 BUG_ON(father
!= p
->real_parent
&& !ptrace
);
711 if (father
== p
->real_parent
) {
712 /* reparent with a reaper, real father it's us */
713 choose_new_parent(p
, reaper
);
714 reparent_thread(p
, father
, 0);
716 /* reparent ptraced task to its real parent */
718 if (p
->exit_state
== EXIT_ZOMBIE
&& p
->exit_signal
!= -1 &&
719 thread_group_empty(p
))
720 do_notify_parent(p
, p
->exit_signal
);
724 * if the ptraced child is a zombie with exit_signal == -1
725 * we must collect it before we exit, or it will remain
726 * zombie forever since we prevented it from self-reap itself
727 * while it was being traced by us, to be able to see it in wait4.
729 if (unlikely(ptrace
&& p
->exit_state
== EXIT_ZOMBIE
&& p
->exit_signal
== -1))
730 list_add(&p
->ptrace_list
, to_release
);
732 list_for_each_safe(_p
, _n
, &father
->ptrace_children
) {
733 p
= list_entry(_p
, struct task_struct
, ptrace_list
);
734 choose_new_parent(p
, reaper
);
735 reparent_thread(p
, father
, 1);
740 * Send signals to all our closest relatives so that they know
741 * to properly mourn us..
743 static void exit_notify(struct task_struct
*tsk
)
746 struct task_struct
*t
;
747 struct list_head ptrace_dead
, *_p
, *_n
;
750 if (signal_pending(tsk
) && !(tsk
->signal
->flags
& SIGNAL_GROUP_EXIT
)
751 && !thread_group_empty(tsk
)) {
753 * This occurs when there was a race between our exit
754 * syscall and a group signal choosing us as the one to
755 * wake up. It could be that we are the only thread
756 * alerted to check for pending signals, but another thread
757 * should be woken now to take the signal since we will not.
758 * Now we'll wake all the threads in the group just to make
759 * sure someone gets all the pending signals.
761 read_lock(&tasklist_lock
);
762 spin_lock_irq(&tsk
->sighand
->siglock
);
763 for (t
= next_thread(tsk
); t
!= tsk
; t
= next_thread(t
))
764 if (!signal_pending(t
) && !(t
->flags
& PF_EXITING
))
765 recalc_sigpending_and_wake(t
);
766 spin_unlock_irq(&tsk
->sighand
->siglock
);
767 read_unlock(&tasklist_lock
);
770 write_lock_irq(&tasklist_lock
);
773 * This does two things:
775 * A. Make init inherit all the child processes
776 * B. Check to see if any process groups have become orphaned
777 * as a result of our exiting, and if they have any stopped
778 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
781 INIT_LIST_HEAD(&ptrace_dead
);
782 forget_original_parent(tsk
, &ptrace_dead
);
783 BUG_ON(!list_empty(&tsk
->children
));
784 BUG_ON(!list_empty(&tsk
->ptrace_children
));
787 * Check to see if any process groups have become orphaned
788 * as a result of our exiting, and if they have any stopped
789 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
791 * Case i: Our father is in a different pgrp than we are
792 * and we were the only connection outside, so our pgrp
793 * is about to become orphaned.
796 t
= tsk
->real_parent
;
798 pgrp
= task_pgrp(tsk
);
799 if ((task_pgrp(t
) != pgrp
) &&
800 (task_session(t
) == task_session(tsk
)) &&
801 will_become_orphaned_pgrp(pgrp
, tsk
) &&
802 has_stopped_jobs(pgrp
)) {
803 __kill_pgrp_info(SIGHUP
, SEND_SIG_PRIV
, pgrp
);
804 __kill_pgrp_info(SIGCONT
, SEND_SIG_PRIV
, pgrp
);
807 /* Let father know we died
809 * Thread signals are configurable, but you aren't going to use
810 * that to send signals to arbitary processes.
811 * That stops right now.
813 * If the parent exec id doesn't match the exec id we saved
814 * when we started then we know the parent has changed security
817 * If our self_exec id doesn't match our parent_exec_id then
818 * we have changed execution domain as these two values started
819 * the same after a fork.
823 if (tsk
->exit_signal
!= SIGCHLD
&& tsk
->exit_signal
!= -1 &&
824 ( tsk
->parent_exec_id
!= t
->self_exec_id
||
825 tsk
->self_exec_id
!= tsk
->parent_exec_id
)
826 && !capable(CAP_KILL
))
827 tsk
->exit_signal
= SIGCHLD
;
830 /* If something other than our normal parent is ptracing us, then
831 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
832 * only has special meaning to our real parent.
834 if (tsk
->exit_signal
!= -1 && thread_group_empty(tsk
)) {
835 int signal
= tsk
->parent
== tsk
->real_parent
? tsk
->exit_signal
: SIGCHLD
;
836 do_notify_parent(tsk
, signal
);
837 } else if (tsk
->ptrace
) {
838 do_notify_parent(tsk
, SIGCHLD
);
842 if (tsk
->exit_signal
== -1 &&
843 (likely(tsk
->ptrace
== 0) ||
844 unlikely(tsk
->parent
->signal
->flags
& SIGNAL_GROUP_EXIT
)))
846 tsk
->exit_state
= state
;
848 write_unlock_irq(&tasklist_lock
);
850 list_for_each_safe(_p
, _n
, &ptrace_dead
) {
852 t
= list_entry(_p
, struct task_struct
, ptrace_list
);
856 /* If the process is dead, release it - nobody will wait for it */
857 if (state
== EXIT_DEAD
)
861 fastcall NORET_TYPE
void do_exit(long code
)
863 struct task_struct
*tsk
= current
;
866 profile_task_exit(tsk
);
868 WARN_ON(atomic_read(&tsk
->fs_excl
));
870 if (unlikely(in_interrupt()))
871 panic("Aiee, killing interrupt handler!");
872 if (unlikely(!tsk
->pid
))
873 panic("Attempted to kill the idle task!");
874 if (unlikely(tsk
== child_reaper(tsk
))) {
875 if (tsk
->nsproxy
->pid_ns
!= &init_pid_ns
)
876 tsk
->nsproxy
->pid_ns
->child_reaper
= init_pid_ns
.child_reaper
;
878 panic("Attempted to kill init!");
882 if (unlikely(current
->ptrace
& PT_TRACE_EXIT
)) {
883 current
->ptrace_message
= code
;
884 ptrace_notify((PTRACE_EVENT_EXIT
<< 8) | SIGTRAP
);
888 * We're taking recursive faults here in do_exit. Safest is to just
889 * leave this task alone and wait for reboot.
891 if (unlikely(tsk
->flags
& PF_EXITING
)) {
893 "Fixing recursive fault but reboot is needed!\n");
895 * We can do this unlocked here. The futex code uses
896 * this flag just to verify whether the pi state
897 * cleanup has been done or not. In the worst case it
898 * loops once more. We pretend that the cleanup was
899 * done as there is no way to return. Either the
900 * OWNER_DIED bit is set by now or we push the blocked
901 * task into the wait for ever nirwana as well.
903 tsk
->flags
|= PF_EXITPIDONE
;
906 set_current_state(TASK_UNINTERRUPTIBLE
);
911 * tsk->flags are checked in the futex code to protect against
912 * an exiting task cleaning up the robust pi futexes.
914 spin_lock_irq(&tsk
->pi_lock
);
915 tsk
->flags
|= PF_EXITING
;
916 spin_unlock_irq(&tsk
->pi_lock
);
918 if (unlikely(in_atomic()))
919 printk(KERN_INFO
"note: %s[%d] exited with preempt_count %d\n",
920 current
->comm
, current
->pid
,
923 acct_update_integrals(tsk
);
925 update_hiwater_rss(tsk
->mm
);
926 update_hiwater_vm(tsk
->mm
);
928 group_dead
= atomic_dec_and_test(&tsk
->signal
->live
);
930 hrtimer_cancel(&tsk
->signal
->real_timer
);
931 exit_itimers(tsk
->signal
);
933 acct_collect(code
, group_dead
);
934 if (unlikely(tsk
->robust_list
))
935 exit_robust_list(tsk
);
936 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
937 if (unlikely(tsk
->compat_robust_list
))
938 compat_exit_robust_list(tsk
);
940 if (unlikely(tsk
->audit_context
))
943 taskstats_exit(tsk
, group_dead
);
956 if (group_dead
&& tsk
->signal
->leader
)
957 disassociate_ctty(1);
959 module_put(task_thread_info(tsk
)->exec_domain
->module
);
961 module_put(tsk
->binfmt
->module
);
963 tsk
->exit_code
= code
;
964 proc_exit_connector(tsk
);
965 exit_task_namespaces(tsk
);
968 mpol_free(tsk
->mempolicy
);
969 tsk
->mempolicy
= NULL
;
972 * This must happen late, after the PID is not
975 if (unlikely(!list_empty(&tsk
->pi_state_list
)))
976 exit_pi_state_list(tsk
);
977 if (unlikely(current
->pi_state_cache
))
978 kfree(current
->pi_state_cache
);
980 * Make sure we are holding no locks:
982 debug_check_no_locks_held(tsk
);
984 * We can do this unlocked here. The futex code uses this flag
985 * just to verify whether the pi state cleanup has been done
986 * or not. In the worst case it loops once more.
988 tsk
->flags
|= PF_EXITPIDONE
;
993 if (tsk
->splice_pipe
)
994 __free_pipe_info(tsk
->splice_pipe
);
997 /* causes final put_task_struct in finish_task_switch(). */
998 tsk
->state
= TASK_DEAD
;
1002 /* Avoid "noreturn function does return". */
1004 cpu_relax(); /* For when BUG is null */
1007 EXPORT_SYMBOL_GPL(do_exit
);
1009 NORET_TYPE
void complete_and_exit(struct completion
*comp
, long code
)
1017 EXPORT_SYMBOL(complete_and_exit
);
1019 asmlinkage
long sys_exit(int error_code
)
1021 do_exit((error_code
&0xff)<<8);
1025 * Take down every thread in the group. This is called by fatal signals
1026 * as well as by sys_exit_group (below).
1029 do_group_exit(int exit_code
)
1031 BUG_ON(exit_code
& 0x80); /* core dumps don't get here */
1033 if (current
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1034 exit_code
= current
->signal
->group_exit_code
;
1035 else if (!thread_group_empty(current
)) {
1036 struct signal_struct
*const sig
= current
->signal
;
1037 struct sighand_struct
*const sighand
= current
->sighand
;
1038 spin_lock_irq(&sighand
->siglock
);
1039 if (sig
->flags
& SIGNAL_GROUP_EXIT
)
1040 /* Another thread got here before we took the lock. */
1041 exit_code
= sig
->group_exit_code
;
1043 sig
->group_exit_code
= exit_code
;
1044 zap_other_threads(current
);
1046 spin_unlock_irq(&sighand
->siglock
);
1054 * this kills every thread in the thread group. Note that any externally
1055 * wait4()-ing process will get the correct exit code - even if this
1056 * thread is not the thread group leader.
1058 asmlinkage
void sys_exit_group(int error_code
)
1060 do_group_exit((error_code
& 0xff) << 8);
1063 static int eligible_child(pid_t pid
, int options
, struct task_struct
*p
)
1071 if (process_group(p
) != process_group(current
))
1073 } else if (pid
!= -1) {
1074 if (process_group(p
) != -pid
)
1079 * Do not consider detached threads that are
1082 if (p
->exit_signal
== -1 && !p
->ptrace
)
1085 /* Wait for all children (clone and not) if __WALL is set;
1086 * otherwise, wait for clone children *only* if __WCLONE is
1087 * set; otherwise, wait for non-clone children *only*. (Note:
1088 * A "clone" child here is one that reports to its parent
1089 * using a signal other than SIGCHLD.) */
1090 if (((p
->exit_signal
!= SIGCHLD
) ^ ((options
& __WCLONE
) != 0))
1091 && !(options
& __WALL
))
1094 * Do not consider thread group leaders that are
1095 * in a non-empty thread group:
1097 if (delay_group_leader(p
))
1100 err
= security_task_wait(p
);
1107 static int wait_noreap_copyout(struct task_struct
*p
, pid_t pid
, uid_t uid
,
1108 int why
, int status
,
1109 struct siginfo __user
*infop
,
1110 struct rusage __user
*rusagep
)
1112 int retval
= rusagep
? getrusage(p
, RUSAGE_BOTH
, rusagep
) : 0;
1116 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1118 retval
= put_user(0, &infop
->si_errno
);
1120 retval
= put_user((short)why
, &infop
->si_code
);
1122 retval
= put_user(pid
, &infop
->si_pid
);
1124 retval
= put_user(uid
, &infop
->si_uid
);
1126 retval
= put_user(status
, &infop
->si_status
);
1133 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1134 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1135 * the lock and this task is uninteresting. If we return nonzero, we have
1136 * released the lock and the system call should return.
1138 static int wait_task_zombie(struct task_struct
*p
, int noreap
,
1139 struct siginfo __user
*infop
,
1140 int __user
*stat_addr
, struct rusage __user
*ru
)
1142 unsigned long state
;
1146 if (unlikely(noreap
)) {
1149 int exit_code
= p
->exit_code
;
1152 if (unlikely(p
->exit_state
!= EXIT_ZOMBIE
))
1154 if (unlikely(p
->exit_signal
== -1 && p
->ptrace
== 0))
1157 read_unlock(&tasklist_lock
);
1158 if ((exit_code
& 0x7f) == 0) {
1160 status
= exit_code
>> 8;
1162 why
= (exit_code
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1163 status
= exit_code
& 0x7f;
1165 return wait_noreap_copyout(p
, pid
, uid
, why
,
1170 * Try to move the task's state to DEAD
1171 * only one thread is allowed to do this:
1173 state
= xchg(&p
->exit_state
, EXIT_DEAD
);
1174 if (state
!= EXIT_ZOMBIE
) {
1175 BUG_ON(state
!= EXIT_DEAD
);
1178 if (unlikely(p
->exit_signal
== -1 && p
->ptrace
== 0)) {
1180 * This can only happen in a race with a ptraced thread
1181 * dying on another processor.
1186 if (likely(p
->real_parent
== p
->parent
) && likely(p
->signal
)) {
1187 struct signal_struct
*psig
;
1188 struct signal_struct
*sig
;
1191 * The resource counters for the group leader are in its
1192 * own task_struct. Those for dead threads in the group
1193 * are in its signal_struct, as are those for the child
1194 * processes it has previously reaped. All these
1195 * accumulate in the parent's signal_struct c* fields.
1197 * We don't bother to take a lock here to protect these
1198 * p->signal fields, because they are only touched by
1199 * __exit_signal, which runs with tasklist_lock
1200 * write-locked anyway, and so is excluded here. We do
1201 * need to protect the access to p->parent->signal fields,
1202 * as other threads in the parent group can be right
1203 * here reaping other children at the same time.
1205 spin_lock_irq(&p
->parent
->sighand
->siglock
);
1206 psig
= p
->parent
->signal
;
1209 cputime_add(psig
->cutime
,
1210 cputime_add(p
->utime
,
1211 cputime_add(sig
->utime
,
1214 cputime_add(psig
->cstime
,
1215 cputime_add(p
->stime
,
1216 cputime_add(sig
->stime
,
1219 p
->min_flt
+ sig
->min_flt
+ sig
->cmin_flt
;
1221 p
->maj_flt
+ sig
->maj_flt
+ sig
->cmaj_flt
;
1223 p
->nvcsw
+ sig
->nvcsw
+ sig
->cnvcsw
;
1225 p
->nivcsw
+ sig
->nivcsw
+ sig
->cnivcsw
;
1227 task_io_get_inblock(p
) +
1228 sig
->inblock
+ sig
->cinblock
;
1230 task_io_get_oublock(p
) +
1231 sig
->oublock
+ sig
->coublock
;
1232 spin_unlock_irq(&p
->parent
->sighand
->siglock
);
1236 * Now we are sure this task is interesting, and no other
1237 * thread can reap it because we set its state to EXIT_DEAD.
1239 read_unlock(&tasklist_lock
);
1241 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1242 status
= (p
->signal
->flags
& SIGNAL_GROUP_EXIT
)
1243 ? p
->signal
->group_exit_code
: p
->exit_code
;
1244 if (!retval
&& stat_addr
)
1245 retval
= put_user(status
, stat_addr
);
1246 if (!retval
&& infop
)
1247 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1248 if (!retval
&& infop
)
1249 retval
= put_user(0, &infop
->si_errno
);
1250 if (!retval
&& infop
) {
1253 if ((status
& 0x7f) == 0) {
1257 why
= (status
& 0x80) ? CLD_DUMPED
: CLD_KILLED
;
1260 retval
= put_user((short)why
, &infop
->si_code
);
1262 retval
= put_user(status
, &infop
->si_status
);
1264 if (!retval
&& infop
)
1265 retval
= put_user(p
->pid
, &infop
->si_pid
);
1266 if (!retval
&& infop
)
1267 retval
= put_user(p
->uid
, &infop
->si_uid
);
1269 // TODO: is this safe?
1270 p
->exit_state
= EXIT_ZOMBIE
;
1274 if (p
->real_parent
!= p
->parent
) {
1275 write_lock_irq(&tasklist_lock
);
1276 /* Double-check with lock held. */
1277 if (p
->real_parent
!= p
->parent
) {
1279 // TODO: is this safe?
1280 p
->exit_state
= EXIT_ZOMBIE
;
1282 * If this is not a detached task, notify the parent.
1283 * If it's still not detached after that, don't release
1286 if (p
->exit_signal
!= -1) {
1287 do_notify_parent(p
, p
->exit_signal
);
1288 if (p
->exit_signal
!= -1)
1292 write_unlock_irq(&tasklist_lock
);
1301 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1302 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1303 * the lock and this task is uninteresting. If we return nonzero, we have
1304 * released the lock and the system call should return.
1306 static int wait_task_stopped(struct task_struct
*p
, int delayed_group_leader
,
1307 int noreap
, struct siginfo __user
*infop
,
1308 int __user
*stat_addr
, struct rusage __user
*ru
)
1310 int retval
, exit_code
;
1314 if (delayed_group_leader
&& !(p
->ptrace
& PT_PTRACED
) &&
1315 p
->signal
&& p
->signal
->group_stop_count
> 0)
1317 * A group stop is in progress and this is the group leader.
1318 * We won't report until all threads have stopped.
1323 * Now we are pretty sure this task is interesting.
1324 * Make sure it doesn't get reaped out from under us while we
1325 * give up the lock and then examine it below. We don't want to
1326 * keep holding onto the tasklist_lock while we call getrusage and
1327 * possibly take page faults for user memory.
1330 read_unlock(&tasklist_lock
);
1332 if (unlikely(noreap
)) {
1335 int why
= (p
->ptrace
& PT_PTRACED
) ? CLD_TRAPPED
: CLD_STOPPED
;
1337 exit_code
= p
->exit_code
;
1338 if (unlikely(!exit_code
) ||
1339 unlikely(p
->state
& TASK_TRACED
))
1341 return wait_noreap_copyout(p
, pid
, uid
,
1342 why
, (exit_code
<< 8) | 0x7f,
1346 write_lock_irq(&tasklist_lock
);
1349 * This uses xchg to be atomic with the thread resuming and setting
1350 * it. It must also be done with the write lock held to prevent a
1351 * race with the EXIT_ZOMBIE case.
1353 exit_code
= xchg(&p
->exit_code
, 0);
1354 if (unlikely(p
->exit_state
)) {
1356 * The task resumed and then died. Let the next iteration
1357 * catch it in EXIT_ZOMBIE. Note that exit_code might
1358 * already be zero here if it resumed and did _exit(0).
1359 * The task itself is dead and won't touch exit_code again;
1360 * other processors in this function are locked out.
1362 p
->exit_code
= exit_code
;
1365 if (unlikely(exit_code
== 0)) {
1367 * Another thread in this function got to it first, or it
1368 * resumed, or it resumed and then died.
1370 write_unlock_irq(&tasklist_lock
);
1374 * We are returning to the wait loop without having successfully
1375 * removed the process and having released the lock. We cannot
1376 * continue, since the "p" task pointer is potentially stale.
1378 * Return -EAGAIN, and do_wait() will restart the loop from the
1379 * beginning. Do _not_ re-acquire the lock.
1384 /* move to end of parent's list to avoid starvation */
1388 write_unlock_irq(&tasklist_lock
);
1390 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1391 if (!retval
&& stat_addr
)
1392 retval
= put_user((exit_code
<< 8) | 0x7f, stat_addr
);
1393 if (!retval
&& infop
)
1394 retval
= put_user(SIGCHLD
, &infop
->si_signo
);
1395 if (!retval
&& infop
)
1396 retval
= put_user(0, &infop
->si_errno
);
1397 if (!retval
&& infop
)
1398 retval
= put_user((short)((p
->ptrace
& PT_PTRACED
)
1399 ? CLD_TRAPPED
: CLD_STOPPED
),
1401 if (!retval
&& infop
)
1402 retval
= put_user(exit_code
, &infop
->si_status
);
1403 if (!retval
&& infop
)
1404 retval
= put_user(p
->pid
, &infop
->si_pid
);
1405 if (!retval
&& infop
)
1406 retval
= put_user(p
->uid
, &infop
->si_uid
);
1416 * Handle do_wait work for one task in a live, non-stopped state.
1417 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1418 * the lock and this task is uninteresting. If we return nonzero, we have
1419 * released the lock and the system call should return.
1421 static int wait_task_continued(struct task_struct
*p
, int noreap
,
1422 struct siginfo __user
*infop
,
1423 int __user
*stat_addr
, struct rusage __user
*ru
)
1429 if (unlikely(!p
->signal
))
1432 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
))
1435 spin_lock_irq(&p
->sighand
->siglock
);
1436 /* Re-check with the lock held. */
1437 if (!(p
->signal
->flags
& SIGNAL_STOP_CONTINUED
)) {
1438 spin_unlock_irq(&p
->sighand
->siglock
);
1442 p
->signal
->flags
&= ~SIGNAL_STOP_CONTINUED
;
1443 spin_unlock_irq(&p
->sighand
->siglock
);
1448 read_unlock(&tasklist_lock
);
1451 retval
= ru
? getrusage(p
, RUSAGE_BOTH
, ru
) : 0;
1453 if (!retval
&& stat_addr
)
1454 retval
= put_user(0xffff, stat_addr
);
1458 retval
= wait_noreap_copyout(p
, pid
, uid
,
1459 CLD_CONTINUED
, SIGCONT
,
1461 BUG_ON(retval
== 0);
1468 static inline int my_ptrace_child(struct task_struct
*p
)
1470 if (!(p
->ptrace
& PT_PTRACED
))
1472 if (!(p
->ptrace
& PT_ATTACHED
))
1475 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1476 * we are the attacher. If we are the real parent, this is a race
1477 * inside ptrace_attach. It is waiting for the tasklist_lock,
1478 * which we have to switch the parent links, but has already set
1479 * the flags in p->ptrace.
1481 return (p
->parent
!= p
->real_parent
);
1484 static long do_wait(pid_t pid
, int options
, struct siginfo __user
*infop
,
1485 int __user
*stat_addr
, struct rusage __user
*ru
)
1487 DECLARE_WAITQUEUE(wait
, current
);
1488 struct task_struct
*tsk
;
1490 int allowed
, denied
;
1492 add_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1495 * We will set this flag if we see any child that might later
1496 * match our criteria, even if we are not able to reap it yet.
1499 allowed
= denied
= 0;
1500 current
->state
= TASK_INTERRUPTIBLE
;
1501 read_lock(&tasklist_lock
);
1504 struct task_struct
*p
;
1505 struct list_head
*_p
;
1508 list_for_each(_p
,&tsk
->children
) {
1509 p
= list_entry(_p
, struct task_struct
, sibling
);
1511 ret
= eligible_child(pid
, options
, p
);
1515 if (unlikely(ret
< 0)) {
1524 * When we hit the race with PTRACE_ATTACH,
1525 * we will not report this child. But the
1526 * race means it has not yet been moved to
1527 * our ptrace_children list, so we need to
1528 * set the flag here to avoid a spurious ECHILD
1529 * when the race happens with the only child.
1532 if (!my_ptrace_child(p
))
1537 * It's stopped now, so it might later
1538 * continue, exit, or stop again.
1541 if (!(options
& WUNTRACED
) &&
1542 !my_ptrace_child(p
))
1544 retval
= wait_task_stopped(p
, ret
== 2,
1545 (options
& WNOWAIT
),
1548 if (retval
== -EAGAIN
)
1550 if (retval
!= 0) /* He released the lock. */
1555 if (p
->exit_state
== EXIT_DEAD
)
1557 // case EXIT_ZOMBIE:
1558 if (p
->exit_state
== EXIT_ZOMBIE
) {
1560 * Eligible but we cannot release
1564 goto check_continued
;
1565 if (!likely(options
& WEXITED
))
1567 retval
= wait_task_zombie(
1568 p
, (options
& WNOWAIT
),
1569 infop
, stat_addr
, ru
);
1570 /* He released the lock. */
1577 * It's running now, so it might later
1578 * exit, stop, or stop and then continue.
1581 if (!unlikely(options
& WCONTINUED
))
1583 retval
= wait_task_continued(
1584 p
, (options
& WNOWAIT
),
1585 infop
, stat_addr
, ru
);
1586 if (retval
!= 0) /* He released the lock. */
1592 list_for_each(_p
, &tsk
->ptrace_children
) {
1593 p
= list_entry(_p
, struct task_struct
,
1595 if (!eligible_child(pid
, options
, p
))
1601 if (options
& __WNOTHREAD
)
1603 tsk
= next_thread(tsk
);
1604 BUG_ON(tsk
->signal
!= current
->signal
);
1605 } while (tsk
!= current
);
1607 read_unlock(&tasklist_lock
);
1610 if (options
& WNOHANG
)
1612 retval
= -ERESTARTSYS
;
1613 if (signal_pending(current
))
1619 if (unlikely(denied
) && !allowed
)
1622 current
->state
= TASK_RUNNING
;
1623 remove_wait_queue(¤t
->signal
->wait_chldexit
,&wait
);
1629 * For a WNOHANG return, clear out all the fields
1630 * we would set so the user can easily tell the
1634 retval
= put_user(0, &infop
->si_signo
);
1636 retval
= put_user(0, &infop
->si_errno
);
1638 retval
= put_user(0, &infop
->si_code
);
1640 retval
= put_user(0, &infop
->si_pid
);
1642 retval
= put_user(0, &infop
->si_uid
);
1644 retval
= put_user(0, &infop
->si_status
);
1650 asmlinkage
long sys_waitid(int which
, pid_t pid
,
1651 struct siginfo __user
*infop
, int options
,
1652 struct rusage __user
*ru
)
1656 if (options
& ~(WNOHANG
|WNOWAIT
|WEXITED
|WSTOPPED
|WCONTINUED
))
1658 if (!(options
& (WEXITED
|WSTOPPED
|WCONTINUED
)))
1678 ret
= do_wait(pid
, options
, infop
, NULL
, ru
);
1680 /* avoid REGPARM breakage on x86: */
1681 prevent_tail_call(ret
);
1685 asmlinkage
long sys_wait4(pid_t pid
, int __user
*stat_addr
,
1686 int options
, struct rusage __user
*ru
)
1690 if (options
& ~(WNOHANG
|WUNTRACED
|WCONTINUED
|
1691 __WNOTHREAD
|__WCLONE
|__WALL
))
1693 ret
= do_wait(pid
, options
| WEXITED
, NULL
, stat_addr
, ru
);
1695 /* avoid REGPARM breakage on x86: */
1696 prevent_tail_call(ret
);
1700 #ifdef __ARCH_WANT_SYS_WAITPID
1703 * sys_waitpid() remains for compatibility. waitpid() should be
1704 * implemented by calling sys_wait4() from libc.a.
1706 asmlinkage
long sys_waitpid(pid_t pid
, int __user
*stat_addr
, int options
)
1708 return sys_wait4(pid
, stat_addr
, options
, NULL
);