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>
50 #include <asm/uaccess.h>
51 #include <asm/unistd.h>
52 #include <asm/pgtable.h>
53 #include <asm/mmu_context.h>
55 static void exit_mm(struct task_struct * tsk);
57 static inline int task_detached(struct task_struct *p)
59 return p->exit_signal == -1;
62 static void __unhash_process(struct task_struct *p)
65 detach_pid(p, PIDTYPE_PID);
66 if (thread_group_leader(p)) {
67 detach_pid(p, PIDTYPE_PGID);
68 detach_pid(p, PIDTYPE_SID);
70 list_del_rcu(&p->tasks);
71 __get_cpu_var(process_counts)--;
73 list_del_rcu(&p->thread_group);
74 list_del_init(&p->sibling);
78 * This function expects the tasklist_lock write-locked.
80 static void __exit_signal(struct task_struct *tsk)
82 struct signal_struct *sig = tsk->signal;
83 struct sighand_struct *sighand;
86 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->utime = cputime_add(sig->utime, tsk->utime);
116 sig->stime = cputime_add(sig->stime, tsk->stime);
117 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
118 sig->min_flt += tsk->min_flt;
119 sig->maj_flt += tsk->maj_flt;
120 sig->nvcsw += tsk->nvcsw;
121 sig->nivcsw += tsk->nivcsw;
122 sig->inblock += task_io_get_inblock(tsk);
123 sig->oublock += task_io_get_oublock(tsk);
124 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
125 sig = NULL; /* Marker for below. */
128 __unhash_process(tsk);
131 * Do this under ->siglock, we can race with another thread
132 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
134 flush_sigqueue(&tsk->pending);
138 spin_unlock(&sighand->siglock);
141 __cleanup_sighand(sighand);
142 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
144 flush_sigqueue(&sig->shared_pending);
145 taskstats_tgid_free(sig);
146 __cleanup_signal(sig);
150 static void delayed_put_task_struct(struct rcu_head *rhp)
152 put_task_struct(container_of(rhp, struct task_struct, rcu));
156 * Do final ptrace-related cleanup of a zombie being reaped.
158 * Called with write_lock(&tasklist_lock) held.
160 static void ptrace_release_task(struct task_struct *p)
162 BUG_ON(!list_empty(&p->ptraced));
164 BUG_ON(!list_empty(&p->ptrace_entry));
167 void release_task(struct task_struct * p)
169 struct task_struct *leader;
172 atomic_dec(&p->user->processes);
174 write_lock_irq(&tasklist_lock);
175 ptrace_release_task(p);
179 * If we are the last non-leader member of the thread
180 * group, and the leader is zombie, then notify the
181 * group leader's parent process. (if it wants notification.)
184 leader = p->group_leader;
185 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
186 BUG_ON(task_detached(leader));
187 do_notify_parent(leader, leader->exit_signal);
189 * If we were the last child thread and the leader has
190 * exited already, and the leader's parent ignores SIGCHLD,
191 * then we are the one who should release the leader.
193 * do_notify_parent() will have marked it self-reaping in
196 zap_leader = task_detached(leader);
199 write_unlock_irq(&tasklist_lock);
201 call_rcu(&p->rcu, delayed_put_task_struct);
204 if (unlikely(zap_leader))
209 * This checks not only the pgrp, but falls back on the pid if no
210 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
213 * The caller must hold rcu lock or the tasklist lock.
215 struct pid *session_of_pgrp(struct pid *pgrp)
217 struct task_struct *p;
218 struct pid *sid = NULL;
220 p = pid_task(pgrp, PIDTYPE_PGID);
222 p = pid_task(pgrp, PIDTYPE_PID);
224 sid = task_session(p);
230 * Determine if a process group is "orphaned", according to the POSIX
231 * definition in 2.2.2.52. Orphaned process groups are not to be affected
232 * by terminal-generated stop signals. Newly orphaned process groups are
233 * to receive a SIGHUP and a SIGCONT.
235 * "I ask you, have you ever known what it is to be an orphan?"
237 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
239 struct task_struct *p;
241 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
242 if ((p == ignored_task) ||
243 (p->exit_state && thread_group_empty(p)) ||
244 is_global_init(p->real_parent))
247 if (task_pgrp(p->real_parent) != pgrp &&
248 task_session(p->real_parent) == task_session(p))
250 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
255 int is_current_pgrp_orphaned(void)
259 read_lock(&tasklist_lock);
260 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
261 read_unlock(&tasklist_lock);
266 static int has_stopped_jobs(struct pid *pgrp)
269 struct task_struct *p;
271 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
272 if (!task_is_stopped(p))
276 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
281 * Check to see if any process groups have become orphaned as
282 * a result of our exiting, and if they have any stopped jobs,
283 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
286 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
288 struct pid *pgrp = task_pgrp(tsk);
289 struct task_struct *ignored_task = tsk;
292 /* exit: our father is in a different pgrp than
293 * we are and we were the only connection outside.
295 parent = tsk->real_parent;
297 /* reparent: our child is in a different pgrp than
298 * we are, and it was the only connection outside.
302 if (task_pgrp(parent) != pgrp &&
303 task_session(parent) == task_session(tsk) &&
304 will_become_orphaned_pgrp(pgrp, ignored_task) &&
305 has_stopped_jobs(pgrp)) {
306 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
307 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
312 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
314 * If a kernel thread is launched as a result of a system call, or if
315 * it ever exits, it should generally reparent itself to kthreadd so it
316 * isn't in the way of other processes and is correctly cleaned up on exit.
318 * The various task state such as scheduling policy and priority may have
319 * been inherited from a user process, so we reset them to sane values here.
321 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
323 static void reparent_to_kthreadd(void)
325 write_lock_irq(&tasklist_lock);
327 ptrace_unlink(current);
328 /* Reparent to init */
329 current->real_parent = current->parent = kthreadd_task;
330 list_move_tail(¤t->sibling, ¤t->real_parent->children);
332 /* Set the exit signal to SIGCHLD so we signal init on exit */
333 current->exit_signal = SIGCHLD;
335 if (task_nice(current) < 0)
336 set_user_nice(current, 0);
340 security_task_reparent_to_init(current);
341 memcpy(current->signal->rlim, init_task.signal->rlim,
342 sizeof(current->signal->rlim));
343 atomic_inc(&(INIT_USER->__count));
344 write_unlock_irq(&tasklist_lock);
345 switch_uid(INIT_USER);
348 void __set_special_pids(struct pid *pid)
350 struct task_struct *curr = current->group_leader;
351 pid_t nr = pid_nr(pid);
353 if (task_session(curr) != pid) {
354 change_pid(curr, PIDTYPE_SID, pid);
355 set_task_session(curr, nr);
357 if (task_pgrp(curr) != pid) {
358 change_pid(curr, PIDTYPE_PGID, pid);
359 set_task_pgrp(curr, nr);
363 static void set_special_pids(struct pid *pid)
365 write_lock_irq(&tasklist_lock);
366 __set_special_pids(pid);
367 write_unlock_irq(&tasklist_lock);
371 * Let kernel threads use this to say that they
372 * allow a certain signal (since daemonize() will
373 * have disabled all of them by default).
375 int allow_signal(int sig)
377 if (!valid_signal(sig) || sig < 1)
380 spin_lock_irq(¤t->sighand->siglock);
381 sigdelset(¤t->blocked, sig);
383 /* Kernel threads handle their own signals.
384 Let the signal code know it'll be handled, so
385 that they don't get converted to SIGKILL or
386 just silently dropped */
387 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
390 spin_unlock_irq(¤t->sighand->siglock);
394 EXPORT_SYMBOL(allow_signal);
396 int disallow_signal(int sig)
398 if (!valid_signal(sig) || sig < 1)
401 spin_lock_irq(¤t->sighand->siglock);
402 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
404 spin_unlock_irq(¤t->sighand->siglock);
408 EXPORT_SYMBOL(disallow_signal);
411 * Put all the gunge required to become a kernel thread without
412 * attached user resources in one place where it belongs.
415 void daemonize(const char *name, ...)
418 struct fs_struct *fs;
421 va_start(args, name);
422 vsnprintf(current->comm, sizeof(current->comm), name, args);
426 * If we were started as result of loading a module, close all of the
427 * user space pages. We don't need them, and if we didn't close them
428 * they would be locked into memory.
432 * We don't want to have TIF_FREEZE set if the system-wide hibernation
433 * or suspend transition begins right now.
435 current->flags |= PF_NOFREEZE;
437 if (current->nsproxy != &init_nsproxy) {
438 get_nsproxy(&init_nsproxy);
439 switch_task_namespaces(current, &init_nsproxy);
441 set_special_pids(&init_struct_pid);
442 proc_clear_tty(current);
444 /* Block and flush all signals */
445 sigfillset(&blocked);
446 sigprocmask(SIG_BLOCK, &blocked, NULL);
447 flush_signals(current);
449 /* Become as one with the init task */
451 exit_fs(current); /* current->fs->count--; */
454 atomic_inc(&fs->count);
457 current->files = init_task.files;
458 atomic_inc(¤t->files->count);
460 reparent_to_kthreadd();
463 EXPORT_SYMBOL(daemonize);
465 static void close_files(struct files_struct * files)
473 * It is safe to dereference the fd table without RCU or
474 * ->file_lock because this is the last reference to the
477 fdt = files_fdtable(files);
481 if (i >= fdt->max_fds)
483 set = fdt->open_fds->fds_bits[j++];
486 struct file * file = xchg(&fdt->fd[i], NULL);
488 filp_close(file, files);
498 struct files_struct *get_files_struct(struct task_struct *task)
500 struct files_struct *files;
505 atomic_inc(&files->count);
511 void put_files_struct(struct files_struct *files)
515 if (atomic_dec_and_test(&files->count)) {
518 * Free the fd and fdset arrays if we expanded them.
519 * If the fdtable was embedded, pass files for freeing
520 * at the end of the RCU grace period. Otherwise,
521 * you can free files immediately.
523 fdt = files_fdtable(files);
524 if (fdt != &files->fdtab)
525 kmem_cache_free(files_cachep, files);
530 void reset_files_struct(struct files_struct *files)
532 struct task_struct *tsk = current;
533 struct files_struct *old;
539 put_files_struct(old);
542 void exit_files(struct task_struct *tsk)
544 struct files_struct * files = tsk->files;
550 put_files_struct(files);
554 void put_fs_struct(struct fs_struct *fs)
556 /* No need to hold fs->lock if we are killing it */
557 if (atomic_dec_and_test(&fs->count)) {
560 if (fs->altroot.dentry)
561 path_put(&fs->altroot);
562 kmem_cache_free(fs_cachep, fs);
566 void exit_fs(struct task_struct *tsk)
568 struct fs_struct * fs = tsk->fs;
578 EXPORT_SYMBOL_GPL(exit_fs);
580 #ifdef CONFIG_MM_OWNER
582 * Task p is exiting and it owned mm, lets find a new owner for it
585 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
588 * If there are other users of the mm and the owner (us) is exiting
589 * we need to find a new owner to take on the responsibility.
593 if (atomic_read(&mm->mm_users) <= 1)
600 void mm_update_next_owner(struct mm_struct *mm)
602 struct task_struct *c, *g, *p = current;
605 if (!mm_need_new_owner(mm, p))
608 read_lock(&tasklist_lock);
610 * Search in the children
612 list_for_each_entry(c, &p->children, sibling) {
614 goto assign_new_owner;
618 * Search in the siblings
620 list_for_each_entry(c, &p->parent->children, sibling) {
622 goto assign_new_owner;
626 * Search through everything else. We should not get
629 do_each_thread(g, c) {
631 goto assign_new_owner;
632 } while_each_thread(g, c);
634 read_unlock(&tasklist_lock);
641 * The task_lock protects c->mm from changing.
642 * We always want mm->owner->mm == mm
646 * Delay read_unlock() till we have the task_lock()
647 * to ensure that c does not slip away underneath us
649 read_unlock(&tasklist_lock);
655 cgroup_mm_owner_callbacks(mm->owner, c);
660 #endif /* CONFIG_MM_OWNER */
663 * Turn us into a lazy TLB process if we
666 static void exit_mm(struct task_struct * tsk)
668 struct mm_struct *mm = tsk->mm;
674 * Serialize with any possible pending coredump.
675 * We must hold mmap_sem around checking core_waiters
676 * and clearing tsk->mm. The core-inducing thread
677 * will increment core_waiters for each thread in the
678 * group with ->mm != NULL.
680 down_read(&mm->mmap_sem);
681 if (mm->core_waiters) {
682 up_read(&mm->mmap_sem);
683 down_write(&mm->mmap_sem);
684 if (!--mm->core_waiters)
685 complete(mm->core_startup_done);
686 up_write(&mm->mmap_sem);
688 wait_for_completion(&mm->core_done);
689 down_read(&mm->mmap_sem);
691 atomic_inc(&mm->mm_count);
692 BUG_ON(mm != tsk->active_mm);
693 /* more a memory barrier than a real lock */
696 up_read(&mm->mmap_sem);
697 enter_lazy_tlb(mm, current);
698 /* We don't want this task to be frozen prematurely */
699 clear_freeze_flag(tsk);
701 mm_update_next_owner(mm);
706 * Detach all tasks we were using ptrace on.
707 * Any that need to be release_task'd are put on the @dead list.
709 * Called with write_lock(&tasklist_lock) held.
711 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
713 struct task_struct *p, *n;
715 list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
718 if (p->exit_state != EXIT_ZOMBIE)
722 * If it's a zombie, our attachedness prevented normal
723 * parent notification or self-reaping. Do notification
724 * now if it would have happened earlier. If it should
725 * reap itself, add it to the @dead list. We can't call
726 * release_task() here because we already hold tasklist_lock.
728 * If it's our own child, there is no notification to do.
730 if (!task_detached(p) && thread_group_empty(p)) {
731 if (!same_thread_group(p->real_parent, parent))
732 do_notify_parent(p, p->exit_signal);
735 if (task_detached(p)) {
737 * Mark it as in the process of being reaped.
739 p->exit_state = EXIT_DEAD;
740 list_add(&p->ptrace_entry, dead);
746 * Finish up exit-time ptrace cleanup.
748 * Called without locks.
750 static void ptrace_exit_finish(struct task_struct *parent,
751 struct list_head *dead)
753 struct task_struct *p, *n;
755 BUG_ON(!list_empty(&parent->ptraced));
757 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
758 list_del_init(&p->ptrace_entry);
763 static void reparent_thread(struct task_struct *p, struct task_struct *father)
765 if (p->pdeath_signal)
766 /* We already hold the tasklist_lock here. */
767 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
769 list_move_tail(&p->sibling, &p->real_parent->children);
771 /* If this is a threaded reparent there is no need to
772 * notify anyone anything has happened.
774 if (same_thread_group(p->real_parent, father))
777 /* We don't want people slaying init. */
778 if (!task_detached(p))
779 p->exit_signal = SIGCHLD;
781 /* If we'd notified the old parent about this child's death,
782 * also notify the new parent.
784 if (!ptrace_reparented(p) &&
785 p->exit_state == EXIT_ZOMBIE &&
786 !task_detached(p) && thread_group_empty(p))
787 do_notify_parent(p, p->exit_signal);
789 kill_orphaned_pgrp(p, father);
793 * When we die, we re-parent all our children.
794 * Try to give them to another thread in our thread
795 * group, and if no such member exists, give it to
796 * the child reaper process (ie "init") in our pid
799 static void forget_original_parent(struct task_struct *father)
801 struct task_struct *p, *n, *reaper = father;
802 LIST_HEAD(ptrace_dead);
804 write_lock_irq(&tasklist_lock);
807 * First clean up ptrace if we were using it.
809 ptrace_exit(father, &ptrace_dead);
812 reaper = next_thread(reaper);
813 if (reaper == father) {
814 reaper = task_child_reaper(father);
817 } while (reaper->flags & PF_EXITING);
819 list_for_each_entry_safe(p, n, &father->children, sibling) {
820 p->real_parent = reaper;
821 if (p->parent == father) {
823 p->parent = p->real_parent;
825 reparent_thread(p, father);
828 write_unlock_irq(&tasklist_lock);
829 BUG_ON(!list_empty(&father->children));
831 ptrace_exit_finish(father, &ptrace_dead);
835 * Send signals to all our closest relatives so that they know
836 * to properly mourn us..
838 static void exit_notify(struct task_struct *tsk, int group_dead)
843 * This does two things:
845 * A. Make init inherit all the child processes
846 * B. Check to see if any process groups have become orphaned
847 * as a result of our exiting, and if they have any stopped
848 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
850 forget_original_parent(tsk);
851 exit_task_namespaces(tsk);
853 write_lock_irq(&tasklist_lock);
855 kill_orphaned_pgrp(tsk->group_leader, NULL);
857 /* Let father know we died
859 * Thread signals are configurable, but you aren't going to use
860 * that to send signals to arbitary processes.
861 * That stops right now.
863 * If the parent exec id doesn't match the exec id we saved
864 * when we started then we know the parent has changed security
867 * If our self_exec id doesn't match our parent_exec_id then
868 * we have changed execution domain as these two values started
869 * the same after a fork.
871 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
872 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
873 tsk->self_exec_id != tsk->parent_exec_id) &&
875 tsk->exit_signal = SIGCHLD;
877 /* If something other than our normal parent is ptracing us, then
878 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
879 * only has special meaning to our real parent.
881 if (!task_detached(tsk) && thread_group_empty(tsk)) {
882 int signal = ptrace_reparented(tsk) ?
883 SIGCHLD : tsk->exit_signal;
884 do_notify_parent(tsk, signal);
885 } else if (tsk->ptrace) {
886 do_notify_parent(tsk, SIGCHLD);
890 if (task_detached(tsk) && likely(!tsk->ptrace))
892 tsk->exit_state = state;
894 /* mt-exec, de_thread() is waiting for us */
895 if (thread_group_leader(tsk) &&
896 tsk->signal->notify_count < 0 &&
897 tsk->signal->group_exit_task)
898 wake_up_process(tsk->signal->group_exit_task);
900 write_unlock_irq(&tasklist_lock);
902 /* If the process is dead, release it - nobody will wait for it */
903 if (state == EXIT_DEAD)
907 #ifdef CONFIG_DEBUG_STACK_USAGE
908 static void check_stack_usage(void)
910 static DEFINE_SPINLOCK(low_water_lock);
911 static int lowest_to_date = THREAD_SIZE;
912 unsigned long *n = end_of_stack(current);
917 free = (unsigned long)n - (unsigned long)end_of_stack(current);
919 if (free >= lowest_to_date)
922 spin_lock(&low_water_lock);
923 if (free < lowest_to_date) {
924 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
926 current->comm, free);
927 lowest_to_date = free;
929 spin_unlock(&low_water_lock);
932 static inline void check_stack_usage(void) {}
935 static inline void exit_child_reaper(struct task_struct *tsk)
937 if (likely(tsk->group_leader != task_child_reaper(tsk)))
940 if (tsk->nsproxy->pid_ns == &init_pid_ns)
941 panic("Attempted to kill init!");
944 * @tsk is the last thread in the 'cgroup-init' and is exiting.
945 * Terminate all remaining processes in the namespace and reap them
946 * before exiting @tsk.
948 * Note that @tsk (last thread of cgroup-init) may not necessarily
949 * be the child-reaper (i.e main thread of cgroup-init) of the
950 * namespace i.e the child_reaper may have already exited.
952 * Even after a child_reaper exits, we let it inherit orphaned children,
953 * because, pid_ns->child_reaper remains valid as long as there is
954 * at least one living sub-thread in the cgroup init.
956 * This living sub-thread of the cgroup-init will be notified when
957 * a child inherited by the 'child-reaper' exits (do_notify_parent()
958 * uses __group_send_sig_info()). Further, when reaping child processes,
959 * do_wait() iterates over children of all living sub threads.
961 * i.e even though 'child_reaper' thread is listed as the parent of the
962 * orphaned children, any living sub-thread in the cgroup-init can
963 * perform the role of the child_reaper.
965 zap_pid_ns_processes(tsk->nsproxy->pid_ns);
968 NORET_TYPE void do_exit(long code)
970 struct task_struct *tsk = current;
973 profile_task_exit(tsk);
975 WARN_ON(atomic_read(&tsk->fs_excl));
977 if (unlikely(in_interrupt()))
978 panic("Aiee, killing interrupt handler!");
979 if (unlikely(!tsk->pid))
980 panic("Attempted to kill the idle task!");
982 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
983 current->ptrace_message = code;
984 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
988 * We're taking recursive faults here in do_exit. Safest is to just
989 * leave this task alone and wait for reboot.
991 if (unlikely(tsk->flags & PF_EXITING)) {
993 "Fixing recursive fault but reboot is needed!\n");
995 * We can do this unlocked here. The futex code uses
996 * this flag just to verify whether the pi state
997 * cleanup has been done or not. In the worst case it
998 * loops once more. We pretend that the cleanup was
999 * done as there is no way to return. Either the
1000 * OWNER_DIED bit is set by now or we push the blocked
1001 * task into the wait for ever nirwana as well.
1003 tsk->flags |= PF_EXITPIDONE;
1004 if (tsk->io_context)
1006 set_current_state(TASK_UNINTERRUPTIBLE);
1010 exit_signals(tsk); /* sets PF_EXITING */
1012 * tsk->flags are checked in the futex code to protect against
1013 * an exiting task cleaning up the robust pi futexes.
1016 spin_unlock_wait(&tsk->pi_lock);
1018 if (unlikely(in_atomic()))
1019 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1020 current->comm, task_pid_nr(current),
1023 acct_update_integrals(tsk);
1025 update_hiwater_rss(tsk->mm);
1026 update_hiwater_vm(tsk->mm);
1028 group_dead = atomic_dec_and_test(&tsk->signal->live);
1030 exit_child_reaper(tsk);
1031 hrtimer_cancel(&tsk->signal->real_timer);
1032 exit_itimers(tsk->signal);
1034 acct_collect(code, group_dead);
1036 if (unlikely(tsk->robust_list))
1037 exit_robust_list(tsk);
1038 #ifdef CONFIG_COMPAT
1039 if (unlikely(tsk->compat_robust_list))
1040 compat_exit_robust_list(tsk);
1045 if (unlikely(tsk->audit_context))
1048 tsk->exit_code = code;
1049 taskstats_exit(tsk, group_dead);
1058 check_stack_usage();
1060 cgroup_exit(tsk, 1);
1063 if (group_dead && tsk->signal->leader)
1064 disassociate_ctty(1);
1066 module_put(task_thread_info(tsk)->exec_domain->module);
1068 module_put(tsk->binfmt->module);
1070 proc_exit_connector(tsk);
1071 exit_notify(tsk, group_dead);
1073 mpol_put(tsk->mempolicy);
1074 tsk->mempolicy = NULL;
1078 * This must happen late, after the PID is not
1081 if (unlikely(!list_empty(&tsk->pi_state_list)))
1082 exit_pi_state_list(tsk);
1083 if (unlikely(current->pi_state_cache))
1084 kfree(current->pi_state_cache);
1087 * Make sure we are holding no locks:
1089 debug_check_no_locks_held(tsk);
1091 * We can do this unlocked here. The futex code uses this flag
1092 * just to verify whether the pi state cleanup has been done
1093 * or not. In the worst case it loops once more.
1095 tsk->flags |= PF_EXITPIDONE;
1097 if (tsk->io_context)
1100 if (tsk->splice_pipe)
1101 __free_pipe_info(tsk->splice_pipe);
1104 /* causes final put_task_struct in finish_task_switch(). */
1105 tsk->state = TASK_DEAD;
1109 /* Avoid "noreturn function does return". */
1111 cpu_relax(); /* For when BUG is null */
1114 EXPORT_SYMBOL_GPL(do_exit);
1116 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1124 EXPORT_SYMBOL(complete_and_exit);
1126 asmlinkage long sys_exit(int error_code)
1128 do_exit((error_code&0xff)<<8);
1132 * Take down every thread in the group. This is called by fatal signals
1133 * as well as by sys_exit_group (below).
1136 do_group_exit(int exit_code)
1138 struct signal_struct *sig = current->signal;
1140 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1142 if (signal_group_exit(sig))
1143 exit_code = sig->group_exit_code;
1144 else if (!thread_group_empty(current)) {
1145 struct sighand_struct *const sighand = current->sighand;
1146 spin_lock_irq(&sighand->siglock);
1147 if (signal_group_exit(sig))
1148 /* Another thread got here before we took the lock. */
1149 exit_code = sig->group_exit_code;
1151 sig->group_exit_code = exit_code;
1152 sig->flags = SIGNAL_GROUP_EXIT;
1153 zap_other_threads(current);
1155 spin_unlock_irq(&sighand->siglock);
1163 * this kills every thread in the thread group. Note that any externally
1164 * wait4()-ing process will get the correct exit code - even if this
1165 * thread is not the thread group leader.
1167 asmlinkage void sys_exit_group(int error_code)
1169 do_group_exit((error_code & 0xff) << 8);
1172 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1174 struct pid *pid = NULL;
1175 if (type == PIDTYPE_PID)
1176 pid = task->pids[type].pid;
1177 else if (type < PIDTYPE_MAX)
1178 pid = task->group_leader->pids[type].pid;
1182 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1183 struct task_struct *p)
1187 if (type < PIDTYPE_MAX) {
1188 if (task_pid_type(p, type) != pid)
1192 /* Wait for all children (clone and not) if __WALL is set;
1193 * otherwise, wait for clone children *only* if __WCLONE is
1194 * set; otherwise, wait for non-clone children *only*. (Note:
1195 * A "clone" child here is one that reports to its parent
1196 * using a signal other than SIGCHLD.) */
1197 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1198 && !(options & __WALL))
1201 err = security_task_wait(p);
1205 if (type != PIDTYPE_PID)
1207 /* This child was explicitly requested, abort */
1208 read_unlock(&tasklist_lock);
1212 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1213 int why, int status,
1214 struct siginfo __user *infop,
1215 struct rusage __user *rusagep)
1217 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1221 retval = put_user(SIGCHLD, &infop->si_signo);
1223 retval = put_user(0, &infop->si_errno);
1225 retval = put_user((short)why, &infop->si_code);
1227 retval = put_user(pid, &infop->si_pid);
1229 retval = put_user(uid, &infop->si_uid);
1231 retval = put_user(status, &infop->si_status);
1238 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1239 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1240 * the lock and this task is uninteresting. If we return nonzero, we have
1241 * released the lock and the system call should return.
1243 static int wait_task_zombie(struct task_struct *p, int options,
1244 struct siginfo __user *infop,
1245 int __user *stat_addr, struct rusage __user *ru)
1247 unsigned long state;
1248 int retval, status, traced;
1249 pid_t pid = task_pid_vnr(p);
1251 if (!likely(options & WEXITED))
1254 if (unlikely(options & WNOWAIT)) {
1256 int exit_code = p->exit_code;
1260 read_unlock(&tasklist_lock);
1261 if ((exit_code & 0x7f) == 0) {
1263 status = exit_code >> 8;
1265 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1266 status = exit_code & 0x7f;
1268 return wait_noreap_copyout(p, pid, uid, why,
1273 * Try to move the task's state to DEAD
1274 * only one thread is allowed to do this:
1276 state = xchg(&p->exit_state, EXIT_DEAD);
1277 if (state != EXIT_ZOMBIE) {
1278 BUG_ON(state != EXIT_DEAD);
1282 traced = ptrace_reparented(p);
1284 if (likely(!traced)) {
1285 struct signal_struct *psig;
1286 struct signal_struct *sig;
1289 * The resource counters for the group leader are in its
1290 * own task_struct. Those for dead threads in the group
1291 * are in its signal_struct, as are those for the child
1292 * processes it has previously reaped. All these
1293 * accumulate in the parent's signal_struct c* fields.
1295 * We don't bother to take a lock here to protect these
1296 * p->signal fields, because they are only touched by
1297 * __exit_signal, which runs with tasklist_lock
1298 * write-locked anyway, and so is excluded here. We do
1299 * need to protect the access to p->parent->signal fields,
1300 * as other threads in the parent group can be right
1301 * here reaping other children at the same time.
1303 spin_lock_irq(&p->parent->sighand->siglock);
1304 psig = p->parent->signal;
1307 cputime_add(psig->cutime,
1308 cputime_add(p->utime,
1309 cputime_add(sig->utime,
1312 cputime_add(psig->cstime,
1313 cputime_add(p->stime,
1314 cputime_add(sig->stime,
1317 cputime_add(psig->cgtime,
1318 cputime_add(p->gtime,
1319 cputime_add(sig->gtime,
1322 p->min_flt + sig->min_flt + sig->cmin_flt;
1324 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1326 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1328 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1330 task_io_get_inblock(p) +
1331 sig->inblock + sig->cinblock;
1333 task_io_get_oublock(p) +
1334 sig->oublock + sig->coublock;
1335 spin_unlock_irq(&p->parent->sighand->siglock);
1339 * Now we are sure this task is interesting, and no other
1340 * thread can reap it because we set its state to EXIT_DEAD.
1342 read_unlock(&tasklist_lock);
1344 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1345 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1346 ? p->signal->group_exit_code : p->exit_code;
1347 if (!retval && stat_addr)
1348 retval = put_user(status, stat_addr);
1349 if (!retval && infop)
1350 retval = put_user(SIGCHLD, &infop->si_signo);
1351 if (!retval && infop)
1352 retval = put_user(0, &infop->si_errno);
1353 if (!retval && infop) {
1356 if ((status & 0x7f) == 0) {
1360 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1363 retval = put_user((short)why, &infop->si_code);
1365 retval = put_user(status, &infop->si_status);
1367 if (!retval && infop)
1368 retval = put_user(pid, &infop->si_pid);
1369 if (!retval && infop)
1370 retval = put_user(p->uid, &infop->si_uid);
1375 write_lock_irq(&tasklist_lock);
1376 /* We dropped tasklist, ptracer could die and untrace */
1379 * If this is not a detached task, notify the parent.
1380 * If it's still not detached after that, don't release
1383 if (!task_detached(p)) {
1384 do_notify_parent(p, p->exit_signal);
1385 if (!task_detached(p)) {
1386 p->exit_state = EXIT_ZOMBIE;
1390 write_unlock_irq(&tasklist_lock);
1399 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1400 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1401 * the lock and this task is uninteresting. If we return nonzero, we have
1402 * released the lock and the system call should return.
1404 static int wait_task_stopped(int ptrace, struct task_struct *p,
1405 int options, struct siginfo __user *infop,
1406 int __user *stat_addr, struct rusage __user *ru)
1408 int retval, exit_code, why;
1409 uid_t uid = 0; /* unneeded, required by compiler */
1412 if (!(options & WUNTRACED))
1416 spin_lock_irq(&p->sighand->siglock);
1418 if (unlikely(!task_is_stopped_or_traced(p)))
1421 if (!ptrace && p->signal->group_stop_count > 0)
1423 * A group stop is in progress and this is the group leader.
1424 * We won't report until all threads have stopped.
1428 exit_code = p->exit_code;
1432 if (!unlikely(options & WNOWAIT))
1437 spin_unlock_irq(&p->sighand->siglock);
1442 * Now we are pretty sure this task is interesting.
1443 * Make sure it doesn't get reaped out from under us while we
1444 * give up the lock and then examine it below. We don't want to
1445 * keep holding onto the tasklist_lock while we call getrusage and
1446 * possibly take page faults for user memory.
1449 pid = task_pid_vnr(p);
1450 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1451 read_unlock(&tasklist_lock);
1453 if (unlikely(options & WNOWAIT))
1454 return wait_noreap_copyout(p, pid, uid,
1458 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1459 if (!retval && stat_addr)
1460 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1461 if (!retval && infop)
1462 retval = put_user(SIGCHLD, &infop->si_signo);
1463 if (!retval && infop)
1464 retval = put_user(0, &infop->si_errno);
1465 if (!retval && infop)
1466 retval = put_user((short)why, &infop->si_code);
1467 if (!retval && infop)
1468 retval = put_user(exit_code, &infop->si_status);
1469 if (!retval && infop)
1470 retval = put_user(pid, &infop->si_pid);
1471 if (!retval && infop)
1472 retval = put_user(uid, &infop->si_uid);
1482 * Handle do_wait work for one task in a live, non-stopped state.
1483 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1484 * the lock and this task is uninteresting. If we return nonzero, we have
1485 * released the lock and the system call should return.
1487 static int wait_task_continued(struct task_struct *p, int options,
1488 struct siginfo __user *infop,
1489 int __user *stat_addr, struct rusage __user *ru)
1495 if (!unlikely(options & WCONTINUED))
1498 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1501 spin_lock_irq(&p->sighand->siglock);
1502 /* Re-check with the lock held. */
1503 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1504 spin_unlock_irq(&p->sighand->siglock);
1507 if (!unlikely(options & WNOWAIT))
1508 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1509 spin_unlock_irq(&p->sighand->siglock);
1511 pid = task_pid_vnr(p);
1514 read_unlock(&tasklist_lock);
1517 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1519 if (!retval && stat_addr)
1520 retval = put_user(0xffff, stat_addr);
1524 retval = wait_noreap_copyout(p, pid, uid,
1525 CLD_CONTINUED, SIGCONT,
1527 BUG_ON(retval == 0);
1534 * Consider @p for a wait by @parent.
1536 * -ECHILD should be in *@notask_error before the first call.
1537 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1538 * Returns zero if the search for a child should continue;
1539 * then *@notask_error is 0 if @p is an eligible child, or still -ECHILD.
1541 static int wait_consider_task(struct task_struct *parent, int ptrace,
1542 struct task_struct *p, int *notask_error,
1543 enum pid_type type, struct pid *pid, int options,
1544 struct siginfo __user *infop,
1545 int __user *stat_addr, struct rusage __user *ru)
1547 int ret = eligible_child(type, pid, options, p);
1551 if (likely(!ptrace) && unlikely(p->ptrace)) {
1553 * This child is hidden by ptrace.
1554 * We aren't allowed to see it now, but eventually we will.
1560 if (p->exit_state == EXIT_DEAD)
1564 * We don't reap group leaders with subthreads.
1566 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1567 return wait_task_zombie(p, options, infop, stat_addr, ru);
1570 * It's stopped or running now, so it might
1571 * later continue, exit, or stop again.
1575 if (task_is_stopped_or_traced(p))
1576 return wait_task_stopped(ptrace, p, options,
1577 infop, stat_addr, ru);
1579 return wait_task_continued(p, options, infop, stat_addr, ru);
1583 * Do the work of do_wait() for one thread in the group, @tsk.
1585 * -ECHILD should be in *@notask_error before the first call.
1586 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1587 * Returns zero if the search for a child should continue; then
1588 * *@notask_error is 0 if there were any eligible children, or still -ECHILD.
1590 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1591 enum pid_type type, struct pid *pid, int options,
1592 struct siginfo __user *infop, int __user *stat_addr,
1593 struct rusage __user *ru)
1595 struct task_struct *p;
1597 list_for_each_entry(p, &tsk->children, sibling) {
1599 * Do not consider detached threads.
1601 if (!task_detached(p)) {
1602 int ret = wait_consider_task(tsk, 0, p, notask_error,
1604 infop, stat_addr, ru);
1613 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1614 enum pid_type type, struct pid *pid, int options,
1615 struct siginfo __user *infop, int __user *stat_addr,
1616 struct rusage __user *ru)
1618 struct task_struct *p;
1621 * Traditionally we see ptrace'd stopped tasks regardless of options.
1623 options |= WUNTRACED;
1625 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1626 int ret = wait_consider_task(tsk, 1, p, notask_error,
1628 infop, stat_addr, ru);
1636 static long do_wait(enum pid_type type, struct pid *pid, int options,
1637 struct siginfo __user *infop, int __user *stat_addr,
1638 struct rusage __user *ru)
1640 DECLARE_WAITQUEUE(wait, current);
1641 struct task_struct *tsk;
1644 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1647 * If there is nothing that can match our critiera just get out.
1648 * We will clear @retval to zero if we see any child that might later
1649 * match our criteria, even if we are not able to reap it yet.
1652 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1655 current->state = TASK_INTERRUPTIBLE;
1656 read_lock(&tasklist_lock);
1659 int tsk_result = do_wait_thread(tsk, &retval,
1661 infop, stat_addr, ru);
1663 tsk_result = ptrace_do_wait(tsk, &retval,
1665 infop, stat_addr, ru);
1668 * tasklist_lock is unlocked and we have a final result.
1670 retval = tsk_result;
1674 if (options & __WNOTHREAD)
1676 tsk = next_thread(tsk);
1677 BUG_ON(tsk->signal != current->signal);
1678 } while (tsk != current);
1679 read_unlock(&tasklist_lock);
1681 if (!retval && !(options & WNOHANG)) {
1682 retval = -ERESTARTSYS;
1683 if (!signal_pending(current)) {
1690 current->state = TASK_RUNNING;
1691 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1697 * For a WNOHANG return, clear out all the fields
1698 * we would set so the user can easily tell the
1702 retval = put_user(0, &infop->si_signo);
1704 retval = put_user(0, &infop->si_errno);
1706 retval = put_user(0, &infop->si_code);
1708 retval = put_user(0, &infop->si_pid);
1710 retval = put_user(0, &infop->si_uid);
1712 retval = put_user(0, &infop->si_status);
1718 asmlinkage long sys_waitid(int which, pid_t upid,
1719 struct siginfo __user *infop, int options,
1720 struct rusage __user *ru)
1722 struct pid *pid = NULL;
1726 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1728 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1741 type = PIDTYPE_PGID;
1749 if (type < PIDTYPE_MAX)
1750 pid = find_get_pid(upid);
1751 ret = do_wait(type, pid, options, infop, NULL, ru);
1754 /* avoid REGPARM breakage on x86: */
1755 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1759 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1760 int options, struct rusage __user *ru)
1762 struct pid *pid = NULL;
1766 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1767 __WNOTHREAD|__WCLONE|__WALL))
1772 else if (upid < 0) {
1773 type = PIDTYPE_PGID;
1774 pid = find_get_pid(-upid);
1775 } else if (upid == 0) {
1776 type = PIDTYPE_PGID;
1777 pid = get_pid(task_pgrp(current));
1778 } else /* upid > 0 */ {
1780 pid = find_get_pid(upid);
1783 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1786 /* avoid REGPARM breakage on x86: */
1787 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1791 #ifdef __ARCH_WANT_SYS_WAITPID
1794 * sys_waitpid() remains for compatibility. waitpid() should be
1795 * implemented by calling sys_wait4() from libc.a.
1797 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1799 return sys_wait4(pid, stat_addr, options, NULL);