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/iocontext.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/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/compat.h>
44 #include <linux/pipe_fs_i.h>
45 #include <linux/audit.h> /* for audit_free() */
46 #include <linux/resource.h>
47 #include <linux/blkdev.h>
48 #include <linux/task_io_accounting_ops.h>
49 #include <linux/tracehook.h>
51 #include <asm/uaccess.h>
52 #include <asm/unistd.h>
53 #include <asm/pgtable.h>
54 #include <asm/mmu_context.h>
56 static void exit_mm(struct task_struct * tsk);
58 static inline int task_detached(struct task_struct *p)
60 return p->exit_signal == -1;
63 static void __unhash_process(struct task_struct *p)
66 detach_pid(p, PIDTYPE_PID);
67 if (thread_group_leader(p)) {
68 detach_pid(p, PIDTYPE_PGID);
69 detach_pid(p, PIDTYPE_SID);
71 list_del_rcu(&p->tasks);
72 __get_cpu_var(process_counts)--;
74 list_del_rcu(&p->thread_group);
75 list_del_init(&p->sibling);
79 * This function expects the tasklist_lock write-locked.
81 static void __exit_signal(struct task_struct *tsk)
83 struct signal_struct *sig = tsk->signal;
84 struct sighand_struct *sighand;
87 BUG_ON(!atomic_read(&sig->count));
89 sighand = rcu_dereference(tsk->sighand);
90 spin_lock(&sighand->siglock);
92 posix_cpu_timers_exit(tsk);
93 if (atomic_dec_and_test(&sig->count))
94 posix_cpu_timers_exit_group(tsk);
97 * If there is any task waiting for the group exit
100 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
101 wake_up_process(sig->group_exit_task);
103 if (tsk == sig->curr_target)
104 sig->curr_target = next_thread(tsk);
106 * Accumulate here the counters for all threads but the
107 * group leader as they die, so they can be added into
108 * the process-wide totals when those are taken.
109 * The group leader stays around as a zombie as long
110 * as there are other threads. When it gets reaped,
111 * the exit.c code will add its counts into these totals.
112 * We won't ever get here for the group leader, since it
113 * will have been the last reference on the signal_struct.
115 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
116 sig->min_flt += tsk->min_flt;
117 sig->maj_flt += tsk->maj_flt;
118 sig->nvcsw += tsk->nvcsw;
119 sig->nivcsw += tsk->nivcsw;
120 sig->inblock += task_io_get_inblock(tsk);
121 sig->oublock += task_io_get_oublock(tsk);
122 task_io_accounting_add(&sig->ioac, &tsk->ioac);
123 sig = NULL; /* Marker for below. */
126 __unhash_process(tsk);
129 * Do this under ->siglock, we can race with another thread
130 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
132 flush_sigqueue(&tsk->pending);
136 spin_unlock(&sighand->siglock);
138 __cleanup_sighand(sighand);
139 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
141 flush_sigqueue(&sig->shared_pending);
142 taskstats_tgid_free(sig);
143 __cleanup_signal(sig);
147 static void delayed_put_task_struct(struct rcu_head *rhp)
149 put_task_struct(container_of(rhp, struct task_struct, rcu));
153 void release_task(struct task_struct * p)
155 struct task_struct *leader;
158 tracehook_prepare_release_task(p);
159 atomic_dec(&p->user->processes);
161 write_lock_irq(&tasklist_lock);
162 tracehook_finish_release_task(p);
166 * If we are the last non-leader member of the thread
167 * group, and the leader is zombie, then notify the
168 * group leader's parent process. (if it wants notification.)
171 leader = p->group_leader;
172 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
173 BUG_ON(task_detached(leader));
174 do_notify_parent(leader, leader->exit_signal);
176 * If we were the last child thread and the leader has
177 * exited already, and the leader's parent ignores SIGCHLD,
178 * then we are the one who should release the leader.
180 * do_notify_parent() will have marked it self-reaping in
183 zap_leader = task_detached(leader);
186 * This maintains the invariant that release_task()
187 * only runs on a task in EXIT_DEAD, just for sanity.
190 leader->exit_state = EXIT_DEAD;
193 write_unlock_irq(&tasklist_lock);
195 call_rcu(&p->rcu, delayed_put_task_struct);
198 if (unlikely(zap_leader))
203 * This checks not only the pgrp, but falls back on the pid if no
204 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
207 * The caller must hold rcu lock or the tasklist lock.
209 struct pid *session_of_pgrp(struct pid *pgrp)
211 struct task_struct *p;
212 struct pid *sid = NULL;
214 p = pid_task(pgrp, PIDTYPE_PGID);
216 p = pid_task(pgrp, PIDTYPE_PID);
218 sid = task_session(p);
224 * Determine if a process group is "orphaned", according to the POSIX
225 * definition in 2.2.2.52. Orphaned process groups are not to be affected
226 * by terminal-generated stop signals. Newly orphaned process groups are
227 * to receive a SIGHUP and a SIGCONT.
229 * "I ask you, have you ever known what it is to be an orphan?"
231 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
233 struct task_struct *p;
235 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
236 if ((p == ignored_task) ||
237 (p->exit_state && thread_group_empty(p)) ||
238 is_global_init(p->real_parent))
241 if (task_pgrp(p->real_parent) != pgrp &&
242 task_session(p->real_parent) == task_session(p))
244 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
249 int is_current_pgrp_orphaned(void)
253 read_lock(&tasklist_lock);
254 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
255 read_unlock(&tasklist_lock);
260 static int has_stopped_jobs(struct pid *pgrp)
263 struct task_struct *p;
265 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
266 if (!task_is_stopped(p))
270 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
275 * Check to see if any process groups have become orphaned as
276 * a result of our exiting, and if they have any stopped jobs,
277 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
280 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
282 struct pid *pgrp = task_pgrp(tsk);
283 struct task_struct *ignored_task = tsk;
286 /* exit: our father is in a different pgrp than
287 * we are and we were the only connection outside.
289 parent = tsk->real_parent;
291 /* reparent: our child is in a different pgrp than
292 * we are, and it was the only connection outside.
296 if (task_pgrp(parent) != pgrp &&
297 task_session(parent) == task_session(tsk) &&
298 will_become_orphaned_pgrp(pgrp, ignored_task) &&
299 has_stopped_jobs(pgrp)) {
300 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
301 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
306 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
308 * If a kernel thread is launched as a result of a system call, or if
309 * it ever exits, it should generally reparent itself to kthreadd so it
310 * isn't in the way of other processes and is correctly cleaned up on exit.
312 * The various task state such as scheduling policy and priority may have
313 * been inherited from a user process, so we reset them to sane values here.
315 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
317 static void reparent_to_kthreadd(void)
319 write_lock_irq(&tasklist_lock);
321 ptrace_unlink(current);
322 /* Reparent to init */
323 current->real_parent = current->parent = kthreadd_task;
324 list_move_tail(¤t->sibling, ¤t->real_parent->children);
326 /* Set the exit signal to SIGCHLD so we signal init on exit */
327 current->exit_signal = SIGCHLD;
329 if (task_nice(current) < 0)
330 set_user_nice(current, 0);
334 security_task_reparent_to_init(current);
335 memcpy(current->signal->rlim, init_task.signal->rlim,
336 sizeof(current->signal->rlim));
337 atomic_inc(&(INIT_USER->__count));
338 write_unlock_irq(&tasklist_lock);
339 switch_uid(INIT_USER);
342 void __set_special_pids(struct pid *pid)
344 struct task_struct *curr = current->group_leader;
345 pid_t nr = pid_nr(pid);
347 if (task_session(curr) != pid) {
348 change_pid(curr, PIDTYPE_SID, pid);
349 set_task_session(curr, nr);
351 if (task_pgrp(curr) != pid) {
352 change_pid(curr, PIDTYPE_PGID, pid);
353 set_task_pgrp(curr, nr);
357 static void set_special_pids(struct pid *pid)
359 write_lock_irq(&tasklist_lock);
360 __set_special_pids(pid);
361 write_unlock_irq(&tasklist_lock);
365 * Let kernel threads use this to say that they
366 * allow a certain signal (since daemonize() will
367 * have disabled all of them by default).
369 int allow_signal(int sig)
371 if (!valid_signal(sig) || sig < 1)
374 spin_lock_irq(¤t->sighand->siglock);
375 sigdelset(¤t->blocked, sig);
377 /* Kernel threads handle their own signals.
378 Let the signal code know it'll be handled, so
379 that they don't get converted to SIGKILL or
380 just silently dropped */
381 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
384 spin_unlock_irq(¤t->sighand->siglock);
388 EXPORT_SYMBOL(allow_signal);
390 int disallow_signal(int sig)
392 if (!valid_signal(sig) || sig < 1)
395 spin_lock_irq(¤t->sighand->siglock);
396 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
398 spin_unlock_irq(¤t->sighand->siglock);
402 EXPORT_SYMBOL(disallow_signal);
405 * Put all the gunge required to become a kernel thread without
406 * attached user resources in one place where it belongs.
409 void daemonize(const char *name, ...)
412 struct fs_struct *fs;
415 va_start(args, name);
416 vsnprintf(current->comm, sizeof(current->comm), name, args);
420 * If we were started as result of loading a module, close all of the
421 * user space pages. We don't need them, and if we didn't close them
422 * they would be locked into memory.
426 * We don't want to have TIF_FREEZE set if the system-wide hibernation
427 * or suspend transition begins right now.
429 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
431 if (current->nsproxy != &init_nsproxy) {
432 get_nsproxy(&init_nsproxy);
433 switch_task_namespaces(current, &init_nsproxy);
435 set_special_pids(&init_struct_pid);
436 proc_clear_tty(current);
438 /* Block and flush all signals */
439 sigfillset(&blocked);
440 sigprocmask(SIG_BLOCK, &blocked, NULL);
441 flush_signals(current);
443 /* Become as one with the init task */
445 exit_fs(current); /* current->fs->count--; */
448 atomic_inc(&fs->count);
451 current->files = init_task.files;
452 atomic_inc(¤t->files->count);
454 reparent_to_kthreadd();
457 EXPORT_SYMBOL(daemonize);
459 static void close_files(struct files_struct * files)
467 * It is safe to dereference the fd table without RCU or
468 * ->file_lock because this is the last reference to the
471 fdt = files_fdtable(files);
475 if (i >= fdt->max_fds)
477 set = fdt->open_fds->fds_bits[j++];
480 struct file * file = xchg(&fdt->fd[i], NULL);
482 filp_close(file, files);
492 struct files_struct *get_files_struct(struct task_struct *task)
494 struct files_struct *files;
499 atomic_inc(&files->count);
505 void put_files_struct(struct files_struct *files)
509 if (atomic_dec_and_test(&files->count)) {
512 * Free the fd and fdset arrays if we expanded them.
513 * If the fdtable was embedded, pass files for freeing
514 * at the end of the RCU grace period. Otherwise,
515 * you can free files immediately.
517 fdt = files_fdtable(files);
518 if (fdt != &files->fdtab)
519 kmem_cache_free(files_cachep, files);
524 void reset_files_struct(struct files_struct *files)
526 struct task_struct *tsk = current;
527 struct files_struct *old;
533 put_files_struct(old);
536 void exit_files(struct task_struct *tsk)
538 struct files_struct * files = tsk->files;
544 put_files_struct(files);
548 void put_fs_struct(struct fs_struct *fs)
550 /* No need to hold fs->lock if we are killing it */
551 if (atomic_dec_and_test(&fs->count)) {
554 kmem_cache_free(fs_cachep, fs);
558 void exit_fs(struct task_struct *tsk)
560 struct fs_struct * fs = tsk->fs;
570 EXPORT_SYMBOL_GPL(exit_fs);
572 #ifdef CONFIG_MM_OWNER
574 * Task p is exiting and it owned mm, lets find a new owner for it
577 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
580 * If there are other users of the mm and the owner (us) is exiting
581 * we need to find a new owner to take on the responsibility.
585 if (atomic_read(&mm->mm_users) <= 1)
592 void mm_update_next_owner(struct mm_struct *mm)
594 struct task_struct *c, *g, *p = current;
597 if (!mm_need_new_owner(mm, p))
600 read_lock(&tasklist_lock);
602 * Search in the children
604 list_for_each_entry(c, &p->children, sibling) {
606 goto assign_new_owner;
610 * Search in the siblings
612 list_for_each_entry(c, &p->parent->children, sibling) {
614 goto assign_new_owner;
618 * Search through everything else. We should not get
621 do_each_thread(g, c) {
623 goto assign_new_owner;
624 } while_each_thread(g, c);
626 read_unlock(&tasklist_lock);
633 * The task_lock protects c->mm from changing.
634 * We always want mm->owner->mm == mm
638 * Delay read_unlock() till we have the task_lock()
639 * to ensure that c does not slip away underneath us
641 read_unlock(&tasklist_lock);
647 cgroup_mm_owner_callbacks(mm->owner, c);
652 #endif /* CONFIG_MM_OWNER */
655 * Turn us into a lazy TLB process if we
658 static void exit_mm(struct task_struct * tsk)
660 struct mm_struct *mm = tsk->mm;
661 struct core_state *core_state;
667 * Serialize with any possible pending coredump.
668 * We must hold mmap_sem around checking core_state
669 * and clearing tsk->mm. The core-inducing thread
670 * will increment ->nr_threads for each thread in the
671 * group with ->mm != NULL.
673 down_read(&mm->mmap_sem);
674 core_state = mm->core_state;
676 struct core_thread self;
677 up_read(&mm->mmap_sem);
680 self.next = xchg(&core_state->dumper.next, &self);
682 * Implies mb(), the result of xchg() must be visible
683 * to core_state->dumper.
685 if (atomic_dec_and_test(&core_state->nr_threads))
686 complete(&core_state->startup);
689 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
690 if (!self.task) /* see coredump_finish() */
694 __set_task_state(tsk, TASK_RUNNING);
695 down_read(&mm->mmap_sem);
697 atomic_inc(&mm->mm_count);
698 BUG_ON(mm != tsk->active_mm);
699 /* more a memory barrier than a real lock */
702 up_read(&mm->mmap_sem);
703 enter_lazy_tlb(mm, current);
704 /* We don't want this task to be frozen prematurely */
705 clear_freeze_flag(tsk);
707 mm_update_next_owner(mm);
712 * Return nonzero if @parent's children should reap themselves.
714 * Called with write_lock_irq(&tasklist_lock) held.
716 static int ignoring_children(struct task_struct *parent)
719 struct sighand_struct *psig = parent->sighand;
721 spin_lock_irqsave(&psig->siglock, flags);
722 ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
723 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
724 spin_unlock_irqrestore(&psig->siglock, flags);
729 * Detach all tasks we were using ptrace on.
730 * Any that need to be release_task'd are put on the @dead list.
732 * Called with write_lock(&tasklist_lock) held.
734 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
736 struct task_struct *p, *n;
739 list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
742 if (p->exit_state != EXIT_ZOMBIE)
746 * If it's a zombie, our attachedness prevented normal
747 * parent notification or self-reaping. Do notification
748 * now if it would have happened earlier. If it should
749 * reap itself, add it to the @dead list. We can't call
750 * release_task() here because we already hold tasklist_lock.
752 * If it's our own child, there is no notification to do.
753 * But if our normal children self-reap, then this child
754 * was prevented by ptrace and we must reap it now.
756 if (!task_detached(p) && thread_group_empty(p)) {
757 if (!same_thread_group(p->real_parent, parent))
758 do_notify_parent(p, p->exit_signal);
761 ign = ignoring_children(parent);
767 if (task_detached(p)) {
769 * Mark it as in the process of being reaped.
771 p->exit_state = EXIT_DEAD;
772 list_add(&p->ptrace_entry, dead);
778 * Finish up exit-time ptrace cleanup.
780 * Called without locks.
782 static void ptrace_exit_finish(struct task_struct *parent,
783 struct list_head *dead)
785 struct task_struct *p, *n;
787 BUG_ON(!list_empty(&parent->ptraced));
789 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
790 list_del_init(&p->ptrace_entry);
795 static void reparent_thread(struct task_struct *p, struct task_struct *father)
797 if (p->pdeath_signal)
798 /* We already hold the tasklist_lock here. */
799 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
801 list_move_tail(&p->sibling, &p->real_parent->children);
803 /* If this is a threaded reparent there is no need to
804 * notify anyone anything has happened.
806 if (same_thread_group(p->real_parent, father))
809 /* We don't want people slaying init. */
810 if (!task_detached(p))
811 p->exit_signal = SIGCHLD;
813 /* If we'd notified the old parent about this child's death,
814 * also notify the new parent.
816 if (!ptrace_reparented(p) &&
817 p->exit_state == EXIT_ZOMBIE &&
818 !task_detached(p) && thread_group_empty(p))
819 do_notify_parent(p, p->exit_signal);
821 kill_orphaned_pgrp(p, father);
825 * When we die, we re-parent all our children.
826 * Try to give them to another thread in our thread
827 * group, and if no such member exists, give it to
828 * the child reaper process (ie "init") in our pid
831 static struct task_struct *find_new_reaper(struct task_struct *father)
833 struct pid_namespace *pid_ns = task_active_pid_ns(father);
834 struct task_struct *thread;
837 while_each_thread(father, thread) {
838 if (thread->flags & PF_EXITING)
840 if (unlikely(pid_ns->child_reaper == father))
841 pid_ns->child_reaper = thread;
845 if (unlikely(pid_ns->child_reaper == father)) {
846 write_unlock_irq(&tasklist_lock);
847 if (unlikely(pid_ns == &init_pid_ns))
848 panic("Attempted to kill init!");
850 zap_pid_ns_processes(pid_ns);
851 write_lock_irq(&tasklist_lock);
853 * We can not clear ->child_reaper or leave it alone.
854 * There may by stealth EXIT_DEAD tasks on ->children,
855 * forget_original_parent() must move them somewhere.
857 pid_ns->child_reaper = init_pid_ns.child_reaper;
860 return pid_ns->child_reaper;
863 static void forget_original_parent(struct task_struct *father)
865 struct task_struct *p, *n, *reaper;
866 LIST_HEAD(ptrace_dead);
868 write_lock_irq(&tasklist_lock);
869 reaper = find_new_reaper(father);
871 * First clean up ptrace if we were using it.
873 ptrace_exit(father, &ptrace_dead);
875 list_for_each_entry_safe(p, n, &father->children, sibling) {
876 p->real_parent = reaper;
877 if (p->parent == father) {
879 p->parent = p->real_parent;
881 reparent_thread(p, father);
884 write_unlock_irq(&tasklist_lock);
885 BUG_ON(!list_empty(&father->children));
887 ptrace_exit_finish(father, &ptrace_dead);
891 * Send signals to all our closest relatives so that they know
892 * to properly mourn us..
894 static void exit_notify(struct task_struct *tsk, int group_dead)
900 * This does two things:
902 * A. Make init inherit all the child processes
903 * B. Check to see if any process groups have become orphaned
904 * as a result of our exiting, and if they have any stopped
905 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
907 forget_original_parent(tsk);
908 exit_task_namespaces(tsk);
910 write_lock_irq(&tasklist_lock);
912 kill_orphaned_pgrp(tsk->group_leader, NULL);
914 /* Let father know we died
916 * Thread signals are configurable, but you aren't going to use
917 * that to send signals to arbitary processes.
918 * That stops right now.
920 * If the parent exec id doesn't match the exec id we saved
921 * when we started then we know the parent has changed security
924 * If our self_exec id doesn't match our parent_exec_id then
925 * we have changed execution domain as these two values started
926 * the same after a fork.
928 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
929 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
930 tsk->self_exec_id != tsk->parent_exec_id) &&
932 tsk->exit_signal = SIGCHLD;
934 signal = tracehook_notify_death(tsk, &cookie, group_dead);
936 signal = do_notify_parent(tsk, signal);
938 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
940 /* mt-exec, de_thread() is waiting for us */
941 if (thread_group_leader(tsk) &&
942 tsk->signal->group_exit_task &&
943 tsk->signal->notify_count < 0)
944 wake_up_process(tsk->signal->group_exit_task);
946 write_unlock_irq(&tasklist_lock);
948 tracehook_report_death(tsk, signal, cookie, group_dead);
950 /* If the process is dead, release it - nobody will wait for it */
951 if (signal == DEATH_REAP)
955 #ifdef CONFIG_DEBUG_STACK_USAGE
956 static void check_stack_usage(void)
958 static DEFINE_SPINLOCK(low_water_lock);
959 static int lowest_to_date = THREAD_SIZE;
960 unsigned long *n = end_of_stack(current);
965 free = (unsigned long)n - (unsigned long)end_of_stack(current);
967 if (free >= lowest_to_date)
970 spin_lock(&low_water_lock);
971 if (free < lowest_to_date) {
972 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
974 current->comm, free);
975 lowest_to_date = free;
977 spin_unlock(&low_water_lock);
980 static inline void check_stack_usage(void) {}
983 NORET_TYPE void do_exit(long code)
985 struct task_struct *tsk = current;
988 profile_task_exit(tsk);
990 WARN_ON(atomic_read(&tsk->fs_excl));
992 if (unlikely(in_interrupt()))
993 panic("Aiee, killing interrupt handler!");
994 if (unlikely(!tsk->pid))
995 panic("Attempted to kill the idle task!");
997 tracehook_report_exit(&code);
1000 * We're taking recursive faults here in do_exit. Safest is to just
1001 * leave this task alone and wait for reboot.
1003 if (unlikely(tsk->flags & PF_EXITING)) {
1005 "Fixing recursive fault but reboot is needed!\n");
1007 * We can do this unlocked here. The futex code uses
1008 * this flag just to verify whether the pi state
1009 * cleanup has been done or not. In the worst case it
1010 * loops once more. We pretend that the cleanup was
1011 * done as there is no way to return. Either the
1012 * OWNER_DIED bit is set by now or we push the blocked
1013 * task into the wait for ever nirwana as well.
1015 tsk->flags |= PF_EXITPIDONE;
1016 if (tsk->io_context)
1018 set_current_state(TASK_UNINTERRUPTIBLE);
1022 exit_signals(tsk); /* sets PF_EXITING */
1024 * tsk->flags are checked in the futex code to protect against
1025 * an exiting task cleaning up the robust pi futexes.
1028 spin_unlock_wait(&tsk->pi_lock);
1030 if (unlikely(in_atomic()))
1031 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1032 current->comm, task_pid_nr(current),
1035 acct_update_integrals(tsk);
1037 update_hiwater_rss(tsk->mm);
1038 update_hiwater_vm(tsk->mm);
1040 group_dead = atomic_dec_and_test(&tsk->signal->live);
1042 hrtimer_cancel(&tsk->signal->real_timer);
1043 exit_itimers(tsk->signal);
1045 acct_collect(code, group_dead);
1047 if (unlikely(tsk->robust_list))
1048 exit_robust_list(tsk);
1049 #ifdef CONFIG_COMPAT
1050 if (unlikely(tsk->compat_robust_list))
1051 compat_exit_robust_list(tsk);
1056 if (unlikely(tsk->audit_context))
1059 tsk->exit_code = code;
1060 taskstats_exit(tsk, group_dead);
1069 check_stack_usage();
1071 cgroup_exit(tsk, 1);
1074 if (group_dead && tsk->signal->leader)
1075 disassociate_ctty(1);
1077 module_put(task_thread_info(tsk)->exec_domain->module);
1079 module_put(tsk->binfmt->module);
1081 proc_exit_connector(tsk);
1082 exit_notify(tsk, group_dead);
1084 mpol_put(tsk->mempolicy);
1085 tsk->mempolicy = NULL;
1089 * This must happen late, after the PID is not
1092 if (unlikely(!list_empty(&tsk->pi_state_list)))
1093 exit_pi_state_list(tsk);
1094 if (unlikely(current->pi_state_cache))
1095 kfree(current->pi_state_cache);
1098 * Make sure we are holding no locks:
1100 debug_check_no_locks_held(tsk);
1102 * We can do this unlocked here. The futex code uses this flag
1103 * just to verify whether the pi state cleanup has been done
1104 * or not. In the worst case it loops once more.
1106 tsk->flags |= PF_EXITPIDONE;
1108 if (tsk->io_context)
1111 if (tsk->splice_pipe)
1112 __free_pipe_info(tsk->splice_pipe);
1115 /* causes final put_task_struct in finish_task_switch(). */
1116 tsk->state = TASK_DEAD;
1120 /* Avoid "noreturn function does return". */
1122 cpu_relax(); /* For when BUG is null */
1125 EXPORT_SYMBOL_GPL(do_exit);
1127 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1135 EXPORT_SYMBOL(complete_and_exit);
1137 asmlinkage long sys_exit(int error_code)
1139 do_exit((error_code&0xff)<<8);
1143 * Take down every thread in the group. This is called by fatal signals
1144 * as well as by sys_exit_group (below).
1147 do_group_exit(int exit_code)
1149 struct signal_struct *sig = current->signal;
1151 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1153 if (signal_group_exit(sig))
1154 exit_code = sig->group_exit_code;
1155 else if (!thread_group_empty(current)) {
1156 struct sighand_struct *const sighand = current->sighand;
1157 spin_lock_irq(&sighand->siglock);
1158 if (signal_group_exit(sig))
1159 /* Another thread got here before we took the lock. */
1160 exit_code = sig->group_exit_code;
1162 sig->group_exit_code = exit_code;
1163 sig->flags = SIGNAL_GROUP_EXIT;
1164 zap_other_threads(current);
1166 spin_unlock_irq(&sighand->siglock);
1174 * this kills every thread in the thread group. Note that any externally
1175 * wait4()-ing process will get the correct exit code - even if this
1176 * thread is not the thread group leader.
1178 asmlinkage void sys_exit_group(int error_code)
1180 do_group_exit((error_code & 0xff) << 8);
1183 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1185 struct pid *pid = NULL;
1186 if (type == PIDTYPE_PID)
1187 pid = task->pids[type].pid;
1188 else if (type < PIDTYPE_MAX)
1189 pid = task->group_leader->pids[type].pid;
1193 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1194 struct task_struct *p)
1198 if (type < PIDTYPE_MAX) {
1199 if (task_pid_type(p, type) != pid)
1203 /* Wait for all children (clone and not) if __WALL is set;
1204 * otherwise, wait for clone children *only* if __WCLONE is
1205 * set; otherwise, wait for non-clone children *only*. (Note:
1206 * A "clone" child here is one that reports to its parent
1207 * using a signal other than SIGCHLD.) */
1208 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1209 && !(options & __WALL))
1212 err = security_task_wait(p);
1219 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1220 int why, int status,
1221 struct siginfo __user *infop,
1222 struct rusage __user *rusagep)
1224 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1228 retval = put_user(SIGCHLD, &infop->si_signo);
1230 retval = put_user(0, &infop->si_errno);
1232 retval = put_user((short)why, &infop->si_code);
1234 retval = put_user(pid, &infop->si_pid);
1236 retval = put_user(uid, &infop->si_uid);
1238 retval = put_user(status, &infop->si_status);
1245 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1246 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1247 * the lock and this task is uninteresting. If we return nonzero, we have
1248 * released the lock and the system call should return.
1250 static int wait_task_zombie(struct task_struct *p, int options,
1251 struct siginfo __user *infop,
1252 int __user *stat_addr, struct rusage __user *ru)
1254 unsigned long state;
1255 int retval, status, traced;
1256 pid_t pid = task_pid_vnr(p);
1258 if (!likely(options & WEXITED))
1261 if (unlikely(options & WNOWAIT)) {
1263 int exit_code = p->exit_code;
1267 read_unlock(&tasklist_lock);
1268 if ((exit_code & 0x7f) == 0) {
1270 status = exit_code >> 8;
1272 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1273 status = exit_code & 0x7f;
1275 return wait_noreap_copyout(p, pid, uid, why,
1280 * Try to move the task's state to DEAD
1281 * only one thread is allowed to do this:
1283 state = xchg(&p->exit_state, EXIT_DEAD);
1284 if (state != EXIT_ZOMBIE) {
1285 BUG_ON(state != EXIT_DEAD);
1289 traced = ptrace_reparented(p);
1291 if (likely(!traced)) {
1292 struct signal_struct *psig;
1293 struct signal_struct *sig;
1294 struct task_cputime cputime;
1297 * The resource counters for the group leader are in its
1298 * own task_struct. Those for dead threads in the group
1299 * are in its signal_struct, as are those for the child
1300 * processes it has previously reaped. All these
1301 * accumulate in the parent's signal_struct c* fields.
1303 * We don't bother to take a lock here to protect these
1304 * p->signal fields, because they are only touched by
1305 * __exit_signal, which runs with tasklist_lock
1306 * write-locked anyway, and so is excluded here. We do
1307 * need to protect the access to p->parent->signal fields,
1308 * as other threads in the parent group can be right
1309 * here reaping other children at the same time.
1311 * We use thread_group_cputime() to get times for the thread
1312 * group, which consolidates times for all threads in the
1313 * group including the group leader.
1315 spin_lock_irq(&p->parent->sighand->siglock);
1316 psig = p->parent->signal;
1318 thread_group_cputime(p, &cputime);
1320 cputime_add(psig->cutime,
1321 cputime_add(cputime.utime,
1324 cputime_add(psig->cstime,
1325 cputime_add(cputime.stime,
1328 cputime_add(psig->cgtime,
1329 cputime_add(p->gtime,
1330 cputime_add(sig->gtime,
1333 p->min_flt + sig->min_flt + sig->cmin_flt;
1335 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1337 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1339 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1341 task_io_get_inblock(p) +
1342 sig->inblock + sig->cinblock;
1344 task_io_get_oublock(p) +
1345 sig->oublock + sig->coublock;
1346 task_io_accounting_add(&psig->ioac, &p->ioac);
1347 task_io_accounting_add(&psig->ioac, &sig->ioac);
1348 spin_unlock_irq(&p->parent->sighand->siglock);
1352 * Now we are sure this task is interesting, and no other
1353 * thread can reap it because we set its state to EXIT_DEAD.
1355 read_unlock(&tasklist_lock);
1357 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1358 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1359 ? p->signal->group_exit_code : p->exit_code;
1360 if (!retval && stat_addr)
1361 retval = put_user(status, stat_addr);
1362 if (!retval && infop)
1363 retval = put_user(SIGCHLD, &infop->si_signo);
1364 if (!retval && infop)
1365 retval = put_user(0, &infop->si_errno);
1366 if (!retval && infop) {
1369 if ((status & 0x7f) == 0) {
1373 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1376 retval = put_user((short)why, &infop->si_code);
1378 retval = put_user(status, &infop->si_status);
1380 if (!retval && infop)
1381 retval = put_user(pid, &infop->si_pid);
1382 if (!retval && infop)
1383 retval = put_user(p->uid, &infop->si_uid);
1388 write_lock_irq(&tasklist_lock);
1389 /* We dropped tasklist, ptracer could die and untrace */
1392 * If this is not a detached task, notify the parent.
1393 * If it's still not detached after that, don't release
1396 if (!task_detached(p)) {
1397 do_notify_parent(p, p->exit_signal);
1398 if (!task_detached(p)) {
1399 p->exit_state = EXIT_ZOMBIE;
1403 write_unlock_irq(&tasklist_lock);
1412 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1413 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1414 * the lock and this task is uninteresting. If we return nonzero, we have
1415 * released the lock and the system call should return.
1417 static int wait_task_stopped(int ptrace, struct task_struct *p,
1418 int options, struct siginfo __user *infop,
1419 int __user *stat_addr, struct rusage __user *ru)
1421 int retval, exit_code, why;
1422 uid_t uid = 0; /* unneeded, required by compiler */
1425 if (!(options & WUNTRACED))
1429 spin_lock_irq(&p->sighand->siglock);
1431 if (unlikely(!task_is_stopped_or_traced(p)))
1434 if (!ptrace && p->signal->group_stop_count > 0)
1436 * A group stop is in progress and this is the group leader.
1437 * We won't report until all threads have stopped.
1441 exit_code = p->exit_code;
1445 if (!unlikely(options & WNOWAIT))
1450 spin_unlock_irq(&p->sighand->siglock);
1455 * Now we are pretty sure this task is interesting.
1456 * Make sure it doesn't get reaped out from under us while we
1457 * give up the lock and then examine it below. We don't want to
1458 * keep holding onto the tasklist_lock while we call getrusage and
1459 * possibly take page faults for user memory.
1462 pid = task_pid_vnr(p);
1463 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1464 read_unlock(&tasklist_lock);
1466 if (unlikely(options & WNOWAIT))
1467 return wait_noreap_copyout(p, pid, uid,
1471 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1472 if (!retval && stat_addr)
1473 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1474 if (!retval && infop)
1475 retval = put_user(SIGCHLD, &infop->si_signo);
1476 if (!retval && infop)
1477 retval = put_user(0, &infop->si_errno);
1478 if (!retval && infop)
1479 retval = put_user((short)why, &infop->si_code);
1480 if (!retval && infop)
1481 retval = put_user(exit_code, &infop->si_status);
1482 if (!retval && infop)
1483 retval = put_user(pid, &infop->si_pid);
1484 if (!retval && infop)
1485 retval = put_user(uid, &infop->si_uid);
1495 * Handle do_wait work for one task in a live, non-stopped state.
1496 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1497 * the lock and this task is uninteresting. If we return nonzero, we have
1498 * released the lock and the system call should return.
1500 static int wait_task_continued(struct task_struct *p, int options,
1501 struct siginfo __user *infop,
1502 int __user *stat_addr, struct rusage __user *ru)
1508 if (!unlikely(options & WCONTINUED))
1511 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1514 spin_lock_irq(&p->sighand->siglock);
1515 /* Re-check with the lock held. */
1516 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1517 spin_unlock_irq(&p->sighand->siglock);
1520 if (!unlikely(options & WNOWAIT))
1521 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1522 spin_unlock_irq(&p->sighand->siglock);
1524 pid = task_pid_vnr(p);
1527 read_unlock(&tasklist_lock);
1530 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1532 if (!retval && stat_addr)
1533 retval = put_user(0xffff, stat_addr);
1537 retval = wait_noreap_copyout(p, pid, uid,
1538 CLD_CONTINUED, SIGCONT,
1540 BUG_ON(retval == 0);
1547 * Consider @p for a wait by @parent.
1549 * -ECHILD should be in *@notask_error before the first call.
1550 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1551 * Returns zero if the search for a child should continue;
1552 * then *@notask_error is 0 if @p is an eligible child,
1553 * or another error from security_task_wait(), or still -ECHILD.
1555 static int wait_consider_task(struct task_struct *parent, int ptrace,
1556 struct task_struct *p, int *notask_error,
1557 enum pid_type type, struct pid *pid, int options,
1558 struct siginfo __user *infop,
1559 int __user *stat_addr, struct rusage __user *ru)
1561 int ret = eligible_child(type, pid, options, p);
1565 if (unlikely(ret < 0)) {
1567 * If we have not yet seen any eligible child,
1568 * then let this error code replace -ECHILD.
1569 * A permission error will give the user a clue
1570 * to look for security policy problems, rather
1571 * than for mysterious wait bugs.
1574 *notask_error = ret;
1577 if (likely(!ptrace) && unlikely(p->ptrace)) {
1579 * This child is hidden by ptrace.
1580 * We aren't allowed to see it now, but eventually we will.
1586 if (p->exit_state == EXIT_DEAD)
1590 * We don't reap group leaders with subthreads.
1592 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1593 return wait_task_zombie(p, options, infop, stat_addr, ru);
1596 * It's stopped or running now, so it might
1597 * later continue, exit, or stop again.
1601 if (task_is_stopped_or_traced(p))
1602 return wait_task_stopped(ptrace, p, options,
1603 infop, stat_addr, ru);
1605 return wait_task_continued(p, options, infop, stat_addr, ru);
1609 * Do the work of do_wait() for one thread in the group, @tsk.
1611 * -ECHILD should be in *@notask_error before the first call.
1612 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1613 * Returns zero if the search for a child should continue; then
1614 * *@notask_error is 0 if there were any eligible children,
1615 * or another error from security_task_wait(), or still -ECHILD.
1617 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1618 enum pid_type type, struct pid *pid, int options,
1619 struct siginfo __user *infop, int __user *stat_addr,
1620 struct rusage __user *ru)
1622 struct task_struct *p;
1624 list_for_each_entry(p, &tsk->children, sibling) {
1626 * Do not consider detached threads.
1628 if (!task_detached(p)) {
1629 int ret = wait_consider_task(tsk, 0, p, notask_error,
1631 infop, stat_addr, ru);
1640 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1641 enum pid_type type, struct pid *pid, int options,
1642 struct siginfo __user *infop, int __user *stat_addr,
1643 struct rusage __user *ru)
1645 struct task_struct *p;
1648 * Traditionally we see ptrace'd stopped tasks regardless of options.
1650 options |= WUNTRACED;
1652 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1653 int ret = wait_consider_task(tsk, 1, p, notask_error,
1655 infop, stat_addr, ru);
1663 static long do_wait(enum pid_type type, struct pid *pid, int options,
1664 struct siginfo __user *infop, int __user *stat_addr,
1665 struct rusage __user *ru)
1667 DECLARE_WAITQUEUE(wait, current);
1668 struct task_struct *tsk;
1671 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1674 * If there is nothing that can match our critiera just get out.
1675 * We will clear @retval to zero if we see any child that might later
1676 * match our criteria, even if we are not able to reap it yet.
1679 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1682 current->state = TASK_INTERRUPTIBLE;
1683 read_lock(&tasklist_lock);
1686 int tsk_result = do_wait_thread(tsk, &retval,
1688 infop, stat_addr, ru);
1690 tsk_result = ptrace_do_wait(tsk, &retval,
1692 infop, stat_addr, ru);
1695 * tasklist_lock is unlocked and we have a final result.
1697 retval = tsk_result;
1701 if (options & __WNOTHREAD)
1703 tsk = next_thread(tsk);
1704 BUG_ON(tsk->signal != current->signal);
1705 } while (tsk != current);
1706 read_unlock(&tasklist_lock);
1708 if (!retval && !(options & WNOHANG)) {
1709 retval = -ERESTARTSYS;
1710 if (!signal_pending(current)) {
1717 current->state = TASK_RUNNING;
1718 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1724 * For a WNOHANG return, clear out all the fields
1725 * we would set so the user can easily tell the
1729 retval = put_user(0, &infop->si_signo);
1731 retval = put_user(0, &infop->si_errno);
1733 retval = put_user(0, &infop->si_code);
1735 retval = put_user(0, &infop->si_pid);
1737 retval = put_user(0, &infop->si_uid);
1739 retval = put_user(0, &infop->si_status);
1745 asmlinkage long sys_waitid(int which, pid_t upid,
1746 struct siginfo __user *infop, int options,
1747 struct rusage __user *ru)
1749 struct pid *pid = NULL;
1753 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1755 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1768 type = PIDTYPE_PGID;
1776 if (type < PIDTYPE_MAX)
1777 pid = find_get_pid(upid);
1778 ret = do_wait(type, pid, options, infop, NULL, ru);
1781 /* avoid REGPARM breakage on x86: */
1782 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1786 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1787 int options, struct rusage __user *ru)
1789 struct pid *pid = NULL;
1793 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1794 __WNOTHREAD|__WCLONE|__WALL))
1799 else if (upid < 0) {
1800 type = PIDTYPE_PGID;
1801 pid = find_get_pid(-upid);
1802 } else if (upid == 0) {
1803 type = PIDTYPE_PGID;
1804 pid = get_pid(task_pgrp(current));
1805 } else /* upid > 0 */ {
1807 pid = find_get_pid(upid);
1810 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1813 /* avoid REGPARM breakage on x86: */
1814 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1818 #ifdef __ARCH_WANT_SYS_WAITPID
1821 * sys_waitpid() remains for compatibility. waitpid() should be
1822 * implemented by calling sys_wait4() from libc.a.
1824 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1826 return sys_wait4(pid, stat_addr, options, NULL);