4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/mnt_namespace.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/binfmts.h>
23 #include <linux/nsproxy.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/ptrace.h>
26 #include <linux/profile.h>
27 #include <linux/mount.h>
28 #include <linux/proc_fs.h>
29 #include <linux/kthread.h>
30 #include <linux/mempolicy.h>
31 #include <linux/taskstats_kern.h>
32 #include <linux/delayacct.h>
33 #include <linux/freezer.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);
87 posix_cpu_timers_exit(tsk);
88 if (atomic_dec_and_test(&sig->count))
89 posix_cpu_timers_exit_group(tsk);
92 * If there is any task waiting for the group exit
95 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
96 wake_up_process(sig->group_exit_task);
98 if (tsk == sig->curr_target)
99 sig->curr_target = next_thread(tsk);
101 * Accumulate here the counters for all threads but the
102 * group leader as they die, so they can be added into
103 * the process-wide totals when those are taken.
104 * The group leader stays around as a zombie as long
105 * as there are other threads. When it gets reaped,
106 * the exit.c code will add its counts into these totals.
107 * We won't ever get here for the group leader, since it
108 * will have been the last reference on the signal_struct.
110 sig->utime = cputime_add(sig->utime, tsk->utime);
111 sig->stime = cputime_add(sig->stime, tsk->stime);
112 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
113 sig->min_flt += tsk->min_flt;
114 sig->maj_flt += tsk->maj_flt;
115 sig->nvcsw += tsk->nvcsw;
116 sig->nivcsw += tsk->nivcsw;
117 sig->inblock += task_io_get_inblock(tsk);
118 sig->oublock += task_io_get_oublock(tsk);
119 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
120 sig = NULL; /* Marker for below. */
123 __unhash_process(tsk);
127 spin_unlock(&sighand->siglock);
130 __cleanup_sighand(sighand);
131 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
132 flush_sigqueue(&tsk->pending);
134 flush_sigqueue(&sig->shared_pending);
135 taskstats_tgid_free(sig);
136 __cleanup_signal(sig);
140 static void delayed_put_task_struct(struct rcu_head *rhp)
142 put_task_struct(container_of(rhp, struct task_struct, rcu));
145 void release_task(struct task_struct * p)
147 struct task_struct *leader;
150 atomic_dec(&p->user->processes);
151 write_lock_irq(&tasklist_lock);
153 BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
157 * If we are the last non-leader member of the thread
158 * group, and the leader is zombie, then notify the
159 * group leader's parent process. (if it wants notification.)
162 leader = p->group_leader;
163 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
164 BUG_ON(leader->exit_signal == -1);
165 do_notify_parent(leader, leader->exit_signal);
167 * If we were the last child thread and the leader has
168 * exited already, and the leader's parent ignores SIGCHLD,
169 * then we are the one who should release the leader.
171 * do_notify_parent() will have marked it self-reaping in
174 zap_leader = (leader->exit_signal == -1);
177 write_unlock_irq(&tasklist_lock);
180 call_rcu(&p->rcu, delayed_put_task_struct);
183 if (unlikely(zap_leader))
188 * This checks not only the pgrp, but falls back on the pid if no
189 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
192 * The caller must hold rcu lock or the tasklist lock.
194 struct pid *session_of_pgrp(struct pid *pgrp)
196 struct task_struct *p;
197 struct pid *sid = NULL;
199 p = pid_task(pgrp, PIDTYPE_PGID);
201 p = pid_task(pgrp, PIDTYPE_PID);
203 sid = task_session(p);
209 * Determine if a process group is "orphaned", according to the POSIX
210 * definition in 2.2.2.52. Orphaned process groups are not to be affected
211 * by terminal-generated stop signals. Newly orphaned process groups are
212 * to receive a SIGHUP and a SIGCONT.
214 * "I ask you, have you ever known what it is to be an orphan?"
216 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
218 struct task_struct *p;
221 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
222 if (p == ignored_task
224 || is_init(p->real_parent))
226 if (task_pgrp(p->real_parent) != pgrp &&
227 task_session(p->real_parent) == task_session(p)) {
231 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
232 return ret; /* (sighing) "Often!" */
235 int is_current_pgrp_orphaned(void)
239 read_lock(&tasklist_lock);
240 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
241 read_unlock(&tasklist_lock);
246 static int has_stopped_jobs(struct pid *pgrp)
249 struct task_struct *p;
251 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
252 if (p->state != TASK_STOPPED)
256 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
261 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
263 * If a kernel thread is launched as a result of a system call, or if
264 * it ever exits, it should generally reparent itself to kthreadd so it
265 * isn't in the way of other processes and is correctly cleaned up on exit.
267 * The various task state such as scheduling policy and priority may have
268 * been inherited from a user process, so we reset them to sane values here.
270 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
272 static void reparent_to_kthreadd(void)
274 write_lock_irq(&tasklist_lock);
276 ptrace_unlink(current);
277 /* Reparent to init */
278 remove_parent(current);
279 current->real_parent = current->parent = kthreadd_task;
282 /* Set the exit signal to SIGCHLD so we signal init on exit */
283 current->exit_signal = SIGCHLD;
285 if (task_nice(current) < 0)
286 set_user_nice(current, 0);
290 security_task_reparent_to_init(current);
291 memcpy(current->signal->rlim, init_task.signal->rlim,
292 sizeof(current->signal->rlim));
293 atomic_inc(&(INIT_USER->__count));
294 write_unlock_irq(&tasklist_lock);
295 switch_uid(INIT_USER);
298 void __set_special_pids(pid_t session, pid_t pgrp)
300 struct task_struct *curr = current->group_leader;
302 if (process_session(curr) != session) {
303 detach_pid(curr, PIDTYPE_SID);
304 set_signal_session(curr->signal, session);
305 attach_pid(curr, PIDTYPE_SID, find_pid(session));
307 if (process_group(curr) != pgrp) {
308 detach_pid(curr, PIDTYPE_PGID);
309 curr->signal->pgrp = pgrp;
310 attach_pid(curr, PIDTYPE_PGID, find_pid(pgrp));
314 static void set_special_pids(pid_t session, pid_t pgrp)
316 write_lock_irq(&tasklist_lock);
317 __set_special_pids(session, pgrp);
318 write_unlock_irq(&tasklist_lock);
322 * Let kernel threads use this to say that they
323 * allow a certain signal (since daemonize() will
324 * have disabled all of them by default).
326 int allow_signal(int sig)
328 if (!valid_signal(sig) || sig < 1)
331 spin_lock_irq(¤t->sighand->siglock);
332 sigdelset(¤t->blocked, sig);
334 /* Kernel threads handle their own signals.
335 Let the signal code know it'll be handled, so
336 that they don't get converted to SIGKILL or
337 just silently dropped */
338 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
341 spin_unlock_irq(¤t->sighand->siglock);
345 EXPORT_SYMBOL(allow_signal);
347 int disallow_signal(int sig)
349 if (!valid_signal(sig) || sig < 1)
352 spin_lock_irq(¤t->sighand->siglock);
353 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
355 spin_unlock_irq(¤t->sighand->siglock);
359 EXPORT_SYMBOL(disallow_signal);
362 * Put all the gunge required to become a kernel thread without
363 * attached user resources in one place where it belongs.
366 void daemonize(const char *name, ...)
369 struct fs_struct *fs;
372 va_start(args, name);
373 vsnprintf(current->comm, sizeof(current->comm), name, args);
377 * If we were started as result of loading a module, close all of the
378 * user space pages. We don't need them, and if we didn't close them
379 * they would be locked into memory.
383 * We don't want to have TIF_FREEZE set if the system-wide hibernation
384 * or suspend transition begins right now.
386 current->flags |= PF_NOFREEZE;
388 set_special_pids(1, 1);
389 proc_clear_tty(current);
391 /* Block and flush all signals */
392 sigfillset(&blocked);
393 sigprocmask(SIG_BLOCK, &blocked, NULL);
394 flush_signals(current);
396 /* Become as one with the init task */
398 exit_fs(current); /* current->fs->count--; */
401 atomic_inc(&fs->count);
403 exit_task_namespaces(current);
404 current->nsproxy = init_task.nsproxy;
405 get_task_namespaces(current);
408 current->files = init_task.files;
409 atomic_inc(¤t->files->count);
411 reparent_to_kthreadd();
414 EXPORT_SYMBOL(daemonize);
416 static void close_files(struct files_struct * files)
424 * It is safe to dereference the fd table without RCU or
425 * ->file_lock because this is the last reference to the
428 fdt = files_fdtable(files);
432 if (i >= fdt->max_fds)
434 set = fdt->open_fds->fds_bits[j++];
437 struct file * file = xchg(&fdt->fd[i], NULL);
439 filp_close(file, files);
449 struct files_struct *get_files_struct(struct task_struct *task)
451 struct files_struct *files;
456 atomic_inc(&files->count);
462 void fastcall put_files_struct(struct files_struct *files)
466 if (atomic_dec_and_test(&files->count)) {
469 * Free the fd and fdset arrays if we expanded them.
470 * If the fdtable was embedded, pass files for freeing
471 * at the end of the RCU grace period. Otherwise,
472 * you can free files immediately.
474 fdt = files_fdtable(files);
475 if (fdt != &files->fdtab)
476 kmem_cache_free(files_cachep, files);
481 EXPORT_SYMBOL(put_files_struct);
483 void reset_files_struct(struct task_struct *tsk, struct files_struct *files)
485 struct files_struct *old;
491 put_files_struct(old);
493 EXPORT_SYMBOL(reset_files_struct);
495 static inline void __exit_files(struct task_struct *tsk)
497 struct files_struct * files = tsk->files;
503 put_files_struct(files);
507 void exit_files(struct task_struct *tsk)
512 static inline void __put_fs_struct(struct fs_struct *fs)
514 /* No need to hold fs->lock if we are killing it */
515 if (atomic_dec_and_test(&fs->count)) {
522 mntput(fs->altrootmnt);
524 kmem_cache_free(fs_cachep, fs);
528 void put_fs_struct(struct fs_struct *fs)
533 static inline void __exit_fs(struct task_struct *tsk)
535 struct fs_struct * fs = tsk->fs;
545 void exit_fs(struct task_struct *tsk)
550 EXPORT_SYMBOL_GPL(exit_fs);
553 * Turn us into a lazy TLB process if we
556 static void exit_mm(struct task_struct * tsk)
558 struct mm_struct *mm = tsk->mm;
564 * Serialize with any possible pending coredump.
565 * We must hold mmap_sem around checking core_waiters
566 * and clearing tsk->mm. The core-inducing thread
567 * will increment core_waiters for each thread in the
568 * group with ->mm != NULL.
570 down_read(&mm->mmap_sem);
571 if (mm->core_waiters) {
572 up_read(&mm->mmap_sem);
573 down_write(&mm->mmap_sem);
574 if (!--mm->core_waiters)
575 complete(mm->core_startup_done);
576 up_write(&mm->mmap_sem);
578 wait_for_completion(&mm->core_done);
579 down_read(&mm->mmap_sem);
581 atomic_inc(&mm->mm_count);
582 BUG_ON(mm != tsk->active_mm);
583 /* more a memory barrier than a real lock */
586 up_read(&mm->mmap_sem);
587 enter_lazy_tlb(mm, current);
588 /* We don't want this task to be frozen prematurely */
589 clear_freeze_flag(tsk);
595 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
597 if (p->pdeath_signal)
598 /* We already hold the tasklist_lock here. */
599 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
601 /* Move the child from its dying parent to the new one. */
602 if (unlikely(traced)) {
603 /* Preserve ptrace links if someone else is tracing this child. */
604 list_del_init(&p->ptrace_list);
605 if (p->parent != p->real_parent)
606 list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
608 /* If this child is being traced, then we're the one tracing it
609 * anyway, so let go of it.
613 p->parent = p->real_parent;
616 if (p->state == TASK_TRACED) {
618 * If it was at a trace stop, turn it into
619 * a normal stop since it's no longer being
626 /* If this is a threaded reparent there is no need to
627 * notify anyone anything has happened.
629 if (p->real_parent->group_leader == father->group_leader)
632 /* We don't want people slaying init. */
633 if (p->exit_signal != -1)
634 p->exit_signal = SIGCHLD;
636 /* If we'd notified the old parent about this child's death,
637 * also notify the new parent.
639 if (!traced && p->exit_state == EXIT_ZOMBIE &&
640 p->exit_signal != -1 && thread_group_empty(p))
641 do_notify_parent(p, p->exit_signal);
644 * process group orphan check
645 * Case ii: Our child is in a different pgrp
646 * than we are, and it was the only connection
647 * outside, so the child pgrp is now orphaned.
649 if ((task_pgrp(p) != task_pgrp(father)) &&
650 (task_session(p) == task_session(father))) {
651 struct pid *pgrp = task_pgrp(p);
653 if (will_become_orphaned_pgrp(pgrp, NULL) &&
654 has_stopped_jobs(pgrp)) {
655 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
656 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
662 * When we die, we re-parent all our children.
663 * Try to give them to another thread in our thread
664 * group, and if no such member exists, give it to
665 * the child reaper process (ie "init") in our pid
669 forget_original_parent(struct task_struct *father, struct list_head *to_release)
671 struct task_struct *p, *reaper = father;
672 struct list_head *_p, *_n;
675 reaper = next_thread(reaper);
676 if (reaper == father) {
677 reaper = child_reaper(father);
680 } while (reaper->exit_state);
683 * There are only two places where our children can be:
685 * - in our child list
686 * - in our ptraced child list
688 * Search them and reparent children.
690 list_for_each_safe(_p, _n, &father->children) {
692 p = list_entry(_p, struct task_struct, sibling);
696 /* if father isn't the real parent, then ptrace must be enabled */
697 BUG_ON(father != p->real_parent && !ptrace);
699 if (father == p->real_parent) {
700 /* reparent with a reaper, real father it's us */
701 p->real_parent = reaper;
702 reparent_thread(p, father, 0);
704 /* reparent ptraced task to its real parent */
706 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
707 thread_group_empty(p))
708 do_notify_parent(p, p->exit_signal);
712 * if the ptraced child is a zombie with exit_signal == -1
713 * we must collect it before we exit, or it will remain
714 * zombie forever since we prevented it from self-reap itself
715 * while it was being traced by us, to be able to see it in wait4.
717 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
718 list_add(&p->ptrace_list, to_release);
720 list_for_each_safe(_p, _n, &father->ptrace_children) {
721 p = list_entry(_p, struct task_struct, ptrace_list);
722 p->real_parent = reaper;
723 reparent_thread(p, father, 1);
728 * Send signals to all our closest relatives so that they know
729 * to properly mourn us..
731 static void exit_notify(struct task_struct *tsk)
734 struct task_struct *t;
735 struct list_head ptrace_dead, *_p, *_n;
738 if (signal_pending(tsk) && !(tsk->signal->flags & SIGNAL_GROUP_EXIT)
739 && !thread_group_empty(tsk)) {
741 * This occurs when there was a race between our exit
742 * syscall and a group signal choosing us as the one to
743 * wake up. It could be that we are the only thread
744 * alerted to check for pending signals, but another thread
745 * should be woken now to take the signal since we will not.
746 * Now we'll wake all the threads in the group just to make
747 * sure someone gets all the pending signals.
749 spin_lock_irq(&tsk->sighand->siglock);
750 for (t = next_thread(tsk); t != tsk; t = next_thread(t))
751 if (!signal_pending(t) && !(t->flags & PF_EXITING))
752 recalc_sigpending_and_wake(t);
753 spin_unlock_irq(&tsk->sighand->siglock);
756 write_lock_irq(&tasklist_lock);
759 * This does two things:
761 * A. Make init inherit all the child processes
762 * B. Check to see if any process groups have become orphaned
763 * as a result of our exiting, and if they have any stopped
764 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
767 INIT_LIST_HEAD(&ptrace_dead);
768 forget_original_parent(tsk, &ptrace_dead);
769 BUG_ON(!list_empty(&tsk->children));
770 BUG_ON(!list_empty(&tsk->ptrace_children));
773 * Check to see if any process groups have become orphaned
774 * as a result of our exiting, and if they have any stopped
775 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
777 * Case i: Our father is in a different pgrp than we are
778 * and we were the only connection outside, so our pgrp
779 * is about to become orphaned.
781 t = tsk->real_parent;
783 pgrp = task_pgrp(tsk);
784 if ((task_pgrp(t) != pgrp) &&
785 (task_session(t) == task_session(tsk)) &&
786 will_become_orphaned_pgrp(pgrp, tsk) &&
787 has_stopped_jobs(pgrp)) {
788 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
789 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
792 /* Let father know we died
794 * Thread signals are configurable, but you aren't going to use
795 * that to send signals to arbitary processes.
796 * That stops right now.
798 * If the parent exec id doesn't match the exec id we saved
799 * when we started then we know the parent has changed security
802 * If our self_exec id doesn't match our parent_exec_id then
803 * we have changed execution domain as these two values started
804 * the same after a fork.
806 if (tsk->exit_signal != SIGCHLD && tsk->exit_signal != -1 &&
807 ( tsk->parent_exec_id != t->self_exec_id ||
808 tsk->self_exec_id != tsk->parent_exec_id)
809 && !capable(CAP_KILL))
810 tsk->exit_signal = SIGCHLD;
813 /* If something other than our normal parent is ptracing us, then
814 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
815 * only has special meaning to our real parent.
817 if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
818 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
819 do_notify_parent(tsk, signal);
820 } else if (tsk->ptrace) {
821 do_notify_parent(tsk, SIGCHLD);
825 if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
827 tsk->exit_state = state;
829 if (thread_group_leader(tsk) &&
830 tsk->signal->notify_count < 0 &&
831 tsk->signal->group_exit_task)
832 wake_up_process(tsk->signal->group_exit_task);
834 write_unlock_irq(&tasklist_lock);
836 list_for_each_safe(_p, _n, &ptrace_dead) {
838 t = list_entry(_p, struct task_struct, ptrace_list);
842 /* If the process is dead, release it - nobody will wait for it */
843 if (state == EXIT_DEAD)
847 #ifdef CONFIG_DEBUG_STACK_USAGE
848 static void check_stack_usage(void)
850 static DEFINE_SPINLOCK(low_water_lock);
851 static int lowest_to_date = THREAD_SIZE;
852 unsigned long *n = end_of_stack(current);
857 free = (unsigned long)n - (unsigned long)end_of_stack(current);
859 if (free >= lowest_to_date)
862 spin_lock(&low_water_lock);
863 if (free < lowest_to_date) {
864 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
866 current->comm, free);
867 lowest_to_date = free;
869 spin_unlock(&low_water_lock);
872 static inline void check_stack_usage(void) {}
875 static inline void exit_child_reaper(struct task_struct *tsk)
877 if (likely(tsk->group_leader != child_reaper(tsk)))
880 panic("Attempted to kill init!");
883 fastcall NORET_TYPE void do_exit(long code)
885 struct task_struct *tsk = current;
888 profile_task_exit(tsk);
890 WARN_ON(atomic_read(&tsk->fs_excl));
892 if (unlikely(in_interrupt()))
893 panic("Aiee, killing interrupt handler!");
894 if (unlikely(!tsk->pid))
895 panic("Attempted to kill the idle task!");
897 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
898 current->ptrace_message = code;
899 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
903 * We're taking recursive faults here in do_exit. Safest is to just
904 * leave this task alone and wait for reboot.
906 if (unlikely(tsk->flags & PF_EXITING)) {
908 "Fixing recursive fault but reboot is needed!\n");
910 * We can do this unlocked here. The futex code uses
911 * this flag just to verify whether the pi state
912 * cleanup has been done or not. In the worst case it
913 * loops once more. We pretend that the cleanup was
914 * done as there is no way to return. Either the
915 * OWNER_DIED bit is set by now or we push the blocked
916 * task into the wait for ever nirwana as well.
918 tsk->flags |= PF_EXITPIDONE;
921 set_current_state(TASK_UNINTERRUPTIBLE);
925 tsk->flags |= PF_EXITING;
927 * tsk->flags are checked in the futex code to protect against
928 * an exiting task cleaning up the robust pi futexes.
931 spin_unlock_wait(&tsk->pi_lock);
933 if (unlikely(in_atomic()))
934 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
935 current->comm, current->pid,
938 acct_update_integrals(tsk);
940 update_hiwater_rss(tsk->mm);
941 update_hiwater_vm(tsk->mm);
943 group_dead = atomic_dec_and_test(&tsk->signal->live);
945 exit_child_reaper(tsk);
946 hrtimer_cancel(&tsk->signal->real_timer);
947 exit_itimers(tsk->signal);
949 acct_collect(code, group_dead);
950 if (unlikely(tsk->robust_list))
951 exit_robust_list(tsk);
952 #if defined(CONFIG_FUTEX) && defined(CONFIG_COMPAT)
953 if (unlikely(tsk->compat_robust_list))
954 compat_exit_robust_list(tsk);
958 if (unlikely(tsk->audit_context))
961 tsk->exit_code = code;
962 taskstats_exit(tsk, group_dead);
976 if (group_dead && tsk->signal->leader)
977 disassociate_ctty(1);
979 module_put(task_thread_info(tsk)->exec_domain->module);
981 module_put(tsk->binfmt->module);
983 proc_exit_connector(tsk);
984 exit_task_namespaces(tsk);
987 mpol_free(tsk->mempolicy);
988 tsk->mempolicy = NULL;
991 * This must happen late, after the PID is not
994 if (unlikely(!list_empty(&tsk->pi_state_list)))
995 exit_pi_state_list(tsk);
996 if (unlikely(current->pi_state_cache))
997 kfree(current->pi_state_cache);
999 * Make sure we are holding no locks:
1001 debug_check_no_locks_held(tsk);
1003 * We can do this unlocked here. The futex code uses this flag
1004 * just to verify whether the pi state cleanup has been done
1005 * or not. In the worst case it loops once more.
1007 tsk->flags |= PF_EXITPIDONE;
1009 if (tsk->io_context)
1012 if (tsk->splice_pipe)
1013 __free_pipe_info(tsk->splice_pipe);
1016 /* causes final put_task_struct in finish_task_switch(). */
1017 tsk->state = TASK_DEAD;
1021 /* Avoid "noreturn function does return". */
1023 cpu_relax(); /* For when BUG is null */
1026 EXPORT_SYMBOL_GPL(do_exit);
1028 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1036 EXPORT_SYMBOL(complete_and_exit);
1038 asmlinkage long sys_exit(int error_code)
1040 do_exit((error_code&0xff)<<8);
1044 * Take down every thread in the group. This is called by fatal signals
1045 * as well as by sys_exit_group (below).
1048 do_group_exit(int exit_code)
1050 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1052 if (current->signal->flags & SIGNAL_GROUP_EXIT)
1053 exit_code = current->signal->group_exit_code;
1054 else if (!thread_group_empty(current)) {
1055 struct signal_struct *const sig = current->signal;
1056 struct sighand_struct *const sighand = current->sighand;
1057 spin_lock_irq(&sighand->siglock);
1058 if (sig->flags & SIGNAL_GROUP_EXIT)
1059 /* Another thread got here before we took the lock. */
1060 exit_code = sig->group_exit_code;
1062 sig->group_exit_code = exit_code;
1063 zap_other_threads(current);
1065 spin_unlock_irq(&sighand->siglock);
1073 * this kills every thread in the thread group. Note that any externally
1074 * wait4()-ing process will get the correct exit code - even if this
1075 * thread is not the thread group leader.
1077 asmlinkage void sys_exit_group(int error_code)
1079 do_group_exit((error_code & 0xff) << 8);
1082 static int eligible_child(pid_t pid, int options, struct task_struct *p)
1090 if (process_group(p) != process_group(current))
1092 } else if (pid != -1) {
1093 if (process_group(p) != -pid)
1098 * Do not consider detached threads that are
1101 if (p->exit_signal == -1 && !p->ptrace)
1104 /* Wait for all children (clone and not) if __WALL is set;
1105 * otherwise, wait for clone children *only* if __WCLONE is
1106 * set; otherwise, wait for non-clone children *only*. (Note:
1107 * A "clone" child here is one that reports to its parent
1108 * using a signal other than SIGCHLD.) */
1109 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1110 && !(options & __WALL))
1113 * Do not consider thread group leaders that are
1114 * in a non-empty thread group:
1116 if (delay_group_leader(p))
1119 err = security_task_wait(p);
1126 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1127 int why, int status,
1128 struct siginfo __user *infop,
1129 struct rusage __user *rusagep)
1131 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1135 retval = put_user(SIGCHLD, &infop->si_signo);
1137 retval = put_user(0, &infop->si_errno);
1139 retval = put_user((short)why, &infop->si_code);
1141 retval = put_user(pid, &infop->si_pid);
1143 retval = put_user(uid, &infop->si_uid);
1145 retval = put_user(status, &infop->si_status);
1152 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1153 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1154 * the lock and this task is uninteresting. If we return nonzero, we have
1155 * released the lock and the system call should return.
1157 static int wait_task_zombie(struct task_struct *p, int noreap,
1158 struct siginfo __user *infop,
1159 int __user *stat_addr, struct rusage __user *ru)
1161 unsigned long state;
1162 int retval, status, traced;
1164 if (unlikely(noreap)) {
1167 int exit_code = p->exit_code;
1170 if (unlikely(p->exit_state != EXIT_ZOMBIE))
1172 if (unlikely(p->exit_signal == -1 && p->ptrace == 0))
1175 read_unlock(&tasklist_lock);
1176 if ((exit_code & 0x7f) == 0) {
1178 status = exit_code >> 8;
1180 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1181 status = exit_code & 0x7f;
1183 return wait_noreap_copyout(p, pid, uid, why,
1188 * Try to move the task's state to DEAD
1189 * only one thread is allowed to do this:
1191 state = xchg(&p->exit_state, EXIT_DEAD);
1192 if (state != EXIT_ZOMBIE) {
1193 BUG_ON(state != EXIT_DEAD);
1197 /* traced means p->ptrace, but not vice versa */
1198 traced = (p->real_parent != p->parent);
1200 if (likely(!traced)) {
1201 struct signal_struct *psig;
1202 struct signal_struct *sig;
1205 * The resource counters for the group leader are in its
1206 * own task_struct. Those for dead threads in the group
1207 * are in its signal_struct, as are those for the child
1208 * processes it has previously reaped. All these
1209 * accumulate in the parent's signal_struct c* fields.
1211 * We don't bother to take a lock here to protect these
1212 * p->signal fields, because they are only touched by
1213 * __exit_signal, which runs with tasklist_lock
1214 * write-locked anyway, and so is excluded here. We do
1215 * need to protect the access to p->parent->signal fields,
1216 * as other threads in the parent group can be right
1217 * here reaping other children at the same time.
1219 spin_lock_irq(&p->parent->sighand->siglock);
1220 psig = p->parent->signal;
1223 cputime_add(psig->cutime,
1224 cputime_add(p->utime,
1225 cputime_add(sig->utime,
1228 cputime_add(psig->cstime,
1229 cputime_add(p->stime,
1230 cputime_add(sig->stime,
1233 cputime_add(psig->cgtime,
1234 cputime_add(p->gtime,
1235 cputime_add(sig->gtime,
1238 p->min_flt + sig->min_flt + sig->cmin_flt;
1240 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1242 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1244 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1246 task_io_get_inblock(p) +
1247 sig->inblock + sig->cinblock;
1249 task_io_get_oublock(p) +
1250 sig->oublock + sig->coublock;
1251 spin_unlock_irq(&p->parent->sighand->siglock);
1255 * Now we are sure this task is interesting, and no other
1256 * thread can reap it because we set its state to EXIT_DEAD.
1258 read_unlock(&tasklist_lock);
1260 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1261 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1262 ? p->signal->group_exit_code : p->exit_code;
1263 if (!retval && stat_addr)
1264 retval = put_user(status, stat_addr);
1265 if (!retval && infop)
1266 retval = put_user(SIGCHLD, &infop->si_signo);
1267 if (!retval && infop)
1268 retval = put_user(0, &infop->si_errno);
1269 if (!retval && infop) {
1272 if ((status & 0x7f) == 0) {
1276 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1279 retval = put_user((short)why, &infop->si_code);
1281 retval = put_user(status, &infop->si_status);
1283 if (!retval && infop)
1284 retval = put_user(p->pid, &infop->si_pid);
1285 if (!retval && infop)
1286 retval = put_user(p->uid, &infop->si_uid);
1291 write_lock_irq(&tasklist_lock);
1292 /* We dropped tasklist, ptracer could die and untrace */
1295 * If this is not a detached task, notify the parent.
1296 * If it's still not detached after that, don't release
1299 if (p->exit_signal != -1) {
1300 do_notify_parent(p, p->exit_signal);
1301 if (p->exit_signal != -1) {
1302 p->exit_state = EXIT_ZOMBIE;
1306 write_unlock_irq(&tasklist_lock);
1315 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1316 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1317 * the lock and this task is uninteresting. If we return nonzero, we have
1318 * released the lock and the system call should return.
1320 static int wait_task_stopped(struct task_struct *p, int delayed_group_leader,
1321 int noreap, struct siginfo __user *infop,
1322 int __user *stat_addr, struct rusage __user *ru)
1324 int retval, exit_code;
1328 if (delayed_group_leader && !(p->ptrace & PT_PTRACED) &&
1329 p->signal->group_stop_count > 0)
1331 * A group stop is in progress and this is the group leader.
1332 * We won't report until all threads have stopped.
1337 * Now we are pretty sure this task is interesting.
1338 * Make sure it doesn't get reaped out from under us while we
1339 * give up the lock and then examine it below. We don't want to
1340 * keep holding onto the tasklist_lock while we call getrusage and
1341 * possibly take page faults for user memory.
1344 read_unlock(&tasklist_lock);
1346 if (unlikely(noreap)) {
1349 int why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1351 exit_code = p->exit_code;
1352 if (unlikely(!exit_code) ||
1353 unlikely(p->state & TASK_TRACED))
1355 return wait_noreap_copyout(p, pid, uid,
1356 why, (exit_code << 8) | 0x7f,
1360 write_lock_irq(&tasklist_lock);
1363 * This uses xchg to be atomic with the thread resuming and setting
1364 * it. It must also be done with the write lock held to prevent a
1365 * race with the EXIT_ZOMBIE case.
1367 exit_code = xchg(&p->exit_code, 0);
1368 if (unlikely(p->exit_state)) {
1370 * The task resumed and then died. Let the next iteration
1371 * catch it in EXIT_ZOMBIE. Note that exit_code might
1372 * already be zero here if it resumed and did _exit(0).
1373 * The task itself is dead and won't touch exit_code again;
1374 * other processors in this function are locked out.
1376 p->exit_code = exit_code;
1379 if (unlikely(exit_code == 0)) {
1381 * Another thread in this function got to it first, or it
1382 * resumed, or it resumed and then died.
1384 write_unlock_irq(&tasklist_lock);
1388 * We are returning to the wait loop without having successfully
1389 * removed the process and having released the lock. We cannot
1390 * continue, since the "p" task pointer is potentially stale.
1392 * Return -EAGAIN, and do_wait() will restart the loop from the
1393 * beginning. Do _not_ re-acquire the lock.
1398 /* move to end of parent's list to avoid starvation */
1402 write_unlock_irq(&tasklist_lock);
1404 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1405 if (!retval && stat_addr)
1406 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1407 if (!retval && infop)
1408 retval = put_user(SIGCHLD, &infop->si_signo);
1409 if (!retval && infop)
1410 retval = put_user(0, &infop->si_errno);
1411 if (!retval && infop)
1412 retval = put_user((short)((p->ptrace & PT_PTRACED)
1413 ? CLD_TRAPPED : CLD_STOPPED),
1415 if (!retval && infop)
1416 retval = put_user(exit_code, &infop->si_status);
1417 if (!retval && infop)
1418 retval = put_user(p->pid, &infop->si_pid);
1419 if (!retval && infop)
1420 retval = put_user(p->uid, &infop->si_uid);
1430 * Handle do_wait work for one task in a live, non-stopped state.
1431 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1432 * the lock and this task is uninteresting. If we return nonzero, we have
1433 * released the lock and the system call should return.
1435 static int wait_task_continued(struct task_struct *p, int noreap,
1436 struct siginfo __user *infop,
1437 int __user *stat_addr, struct rusage __user *ru)
1443 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1446 spin_lock_irq(&p->sighand->siglock);
1447 /* Re-check with the lock held. */
1448 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1449 spin_unlock_irq(&p->sighand->siglock);
1453 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1454 spin_unlock_irq(&p->sighand->siglock);
1459 read_unlock(&tasklist_lock);
1462 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1464 if (!retval && stat_addr)
1465 retval = put_user(0xffff, stat_addr);
1469 retval = wait_noreap_copyout(p, pid, uid,
1470 CLD_CONTINUED, SIGCONT,
1472 BUG_ON(retval == 0);
1479 static inline int my_ptrace_child(struct task_struct *p)
1481 if (!(p->ptrace & PT_PTRACED))
1483 if (!(p->ptrace & PT_ATTACHED))
1486 * This child was PTRACE_ATTACH'd. We should be seeing it only if
1487 * we are the attacher. If we are the real parent, this is a race
1488 * inside ptrace_attach. It is waiting for the tasklist_lock,
1489 * which we have to switch the parent links, but has already set
1490 * the flags in p->ptrace.
1492 return (p->parent != p->real_parent);
1495 static long do_wait(pid_t pid, int options, struct siginfo __user *infop,
1496 int __user *stat_addr, struct rusage __user *ru)
1498 DECLARE_WAITQUEUE(wait, current);
1499 struct task_struct *tsk;
1501 int allowed, denied;
1503 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1506 * We will set this flag if we see any child that might later
1507 * match our criteria, even if we are not able to reap it yet.
1510 allowed = denied = 0;
1511 current->state = TASK_INTERRUPTIBLE;
1512 read_lock(&tasklist_lock);
1515 struct task_struct *p;
1516 struct list_head *_p;
1519 list_for_each(_p,&tsk->children) {
1520 p = list_entry(_p, struct task_struct, sibling);
1522 ret = eligible_child(pid, options, p);
1526 if (unlikely(ret < 0)) {
1535 * When we hit the race with PTRACE_ATTACH,
1536 * we will not report this child. But the
1537 * race means it has not yet been moved to
1538 * our ptrace_children list, so we need to
1539 * set the flag here to avoid a spurious ECHILD
1540 * when the race happens with the only child.
1543 if (!my_ptrace_child(p))
1548 * It's stopped now, so it might later
1549 * continue, exit, or stop again.
1552 if (!(options & WUNTRACED) &&
1553 !my_ptrace_child(p))
1555 retval = wait_task_stopped(p, ret == 2,
1556 (options & WNOWAIT),
1559 if (retval == -EAGAIN)
1561 if (retval != 0) /* He released the lock. */
1566 if (p->exit_state == EXIT_DEAD)
1568 // case EXIT_ZOMBIE:
1569 if (p->exit_state == EXIT_ZOMBIE) {
1571 * Eligible but we cannot release
1575 goto check_continued;
1576 if (!likely(options & WEXITED))
1578 retval = wait_task_zombie(
1579 p, (options & WNOWAIT),
1580 infop, stat_addr, ru);
1581 /* He released the lock. */
1588 * It's running now, so it might later
1589 * exit, stop, or stop and then continue.
1592 if (!unlikely(options & WCONTINUED))
1594 retval = wait_task_continued(
1595 p, (options & WNOWAIT),
1596 infop, stat_addr, ru);
1597 if (retval != 0) /* He released the lock. */
1603 list_for_each(_p, &tsk->ptrace_children) {
1604 p = list_entry(_p, struct task_struct,
1606 if (!eligible_child(pid, options, p))
1612 if (options & __WNOTHREAD)
1614 tsk = next_thread(tsk);
1615 BUG_ON(tsk->signal != current->signal);
1616 } while (tsk != current);
1618 read_unlock(&tasklist_lock);
1621 if (options & WNOHANG)
1623 retval = -ERESTARTSYS;
1624 if (signal_pending(current))
1630 if (unlikely(denied) && !allowed)
1633 current->state = TASK_RUNNING;
1634 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1640 * For a WNOHANG return, clear out all the fields
1641 * we would set so the user can easily tell the
1645 retval = put_user(0, &infop->si_signo);
1647 retval = put_user(0, &infop->si_errno);
1649 retval = put_user(0, &infop->si_code);
1651 retval = put_user(0, &infop->si_pid);
1653 retval = put_user(0, &infop->si_uid);
1655 retval = put_user(0, &infop->si_status);
1661 asmlinkage long sys_waitid(int which, pid_t pid,
1662 struct siginfo __user *infop, int options,
1663 struct rusage __user *ru)
1667 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1669 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1689 ret = do_wait(pid, options, infop, NULL, ru);
1691 /* avoid REGPARM breakage on x86: */
1692 prevent_tail_call(ret);
1696 asmlinkage long sys_wait4(pid_t pid, int __user *stat_addr,
1697 int options, struct rusage __user *ru)
1701 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1702 __WNOTHREAD|__WCLONE|__WALL))
1704 ret = do_wait(pid, options | WEXITED, NULL, stat_addr, ru);
1706 /* avoid REGPARM breakage on x86: */
1707 prevent_tail_call(ret);
1711 #ifdef __ARCH_WANT_SYS_WAITPID
1714 * sys_waitpid() remains for compatibility. waitpid() should be
1715 * implemented by calling sys_wait4() from libc.a.
1717 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1719 return sys_wait4(pid, stat_addr, options, NULL);