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->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 #ifdef CONFIG_TASK_XACCT
125 sig->rchar += tsk->rchar;
126 sig->wchar += tsk->wchar;
127 sig->syscr += tsk->syscr;
128 sig->syscw += tsk->syscw;
129 #endif /* CONFIG_TASK_XACCT */
130 #ifdef CONFIG_TASK_IO_ACCOUNTING
131 sig->ioac.read_bytes += tsk->ioac.read_bytes;
132 sig->ioac.write_bytes += tsk->ioac.write_bytes;
133 sig->ioac.cancelled_write_bytes +=
134 tsk->ioac.cancelled_write_bytes;
135 #endif /* CONFIG_TASK_IO_ACCOUNTING */
136 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
137 sig = NULL; /* Marker for below. */
140 __unhash_process(tsk);
143 * Do this under ->siglock, we can race with another thread
144 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
146 flush_sigqueue(&tsk->pending);
150 spin_unlock(&sighand->siglock);
152 __cleanup_sighand(sighand);
153 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
155 flush_sigqueue(&sig->shared_pending);
156 taskstats_tgid_free(sig);
157 __cleanup_signal(sig);
161 static void delayed_put_task_struct(struct rcu_head *rhp)
163 put_task_struct(container_of(rhp, struct task_struct, rcu));
167 void release_task(struct task_struct * p)
169 struct task_struct *leader;
172 tracehook_prepare_release_task(p);
173 atomic_dec(&p->user->processes);
175 write_lock_irq(&tasklist_lock);
176 tracehook_finish_release_task(p);
180 * If we are the last non-leader member of the thread
181 * group, and the leader is zombie, then notify the
182 * group leader's parent process. (if it wants notification.)
185 leader = p->group_leader;
186 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
187 BUG_ON(task_detached(leader));
188 do_notify_parent(leader, leader->exit_signal);
190 * If we were the last child thread and the leader has
191 * exited already, and the leader's parent ignores SIGCHLD,
192 * then we are the one who should release the leader.
194 * do_notify_parent() will have marked it self-reaping in
197 zap_leader = task_detached(leader);
200 * This maintains the invariant that release_task()
201 * only runs on a task in EXIT_DEAD, just for sanity.
204 leader->exit_state = EXIT_DEAD;
207 write_unlock_irq(&tasklist_lock);
209 call_rcu(&p->rcu, delayed_put_task_struct);
212 if (unlikely(zap_leader))
217 * This checks not only the pgrp, but falls back on the pid if no
218 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
221 * The caller must hold rcu lock or the tasklist lock.
223 struct pid *session_of_pgrp(struct pid *pgrp)
225 struct task_struct *p;
226 struct pid *sid = NULL;
228 p = pid_task(pgrp, PIDTYPE_PGID);
230 p = pid_task(pgrp, PIDTYPE_PID);
232 sid = task_session(p);
238 * Determine if a process group is "orphaned", according to the POSIX
239 * definition in 2.2.2.52. Orphaned process groups are not to be affected
240 * by terminal-generated stop signals. Newly orphaned process groups are
241 * to receive a SIGHUP and a SIGCONT.
243 * "I ask you, have you ever known what it is to be an orphan?"
245 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
247 struct task_struct *p;
249 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
250 if ((p == ignored_task) ||
251 (p->exit_state && thread_group_empty(p)) ||
252 is_global_init(p->real_parent))
255 if (task_pgrp(p->real_parent) != pgrp &&
256 task_session(p->real_parent) == task_session(p))
258 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
263 int is_current_pgrp_orphaned(void)
267 read_lock(&tasklist_lock);
268 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
269 read_unlock(&tasklist_lock);
274 static int has_stopped_jobs(struct pid *pgrp)
277 struct task_struct *p;
279 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
280 if (!task_is_stopped(p))
284 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
289 * Check to see if any process groups have become orphaned as
290 * a result of our exiting, and if they have any stopped jobs,
291 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
294 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
296 struct pid *pgrp = task_pgrp(tsk);
297 struct task_struct *ignored_task = tsk;
300 /* exit: our father is in a different pgrp than
301 * we are and we were the only connection outside.
303 parent = tsk->real_parent;
305 /* reparent: our child is in a different pgrp than
306 * we are, and it was the only connection outside.
310 if (task_pgrp(parent) != pgrp &&
311 task_session(parent) == task_session(tsk) &&
312 will_become_orphaned_pgrp(pgrp, ignored_task) &&
313 has_stopped_jobs(pgrp)) {
314 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
315 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
320 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
322 * If a kernel thread is launched as a result of a system call, or if
323 * it ever exits, it should generally reparent itself to kthreadd so it
324 * isn't in the way of other processes and is correctly cleaned up on exit.
326 * The various task state such as scheduling policy and priority may have
327 * been inherited from a user process, so we reset them to sane values here.
329 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
331 static void reparent_to_kthreadd(void)
333 write_lock_irq(&tasklist_lock);
335 ptrace_unlink(current);
336 /* Reparent to init */
337 current->real_parent = current->parent = kthreadd_task;
338 list_move_tail(¤t->sibling, ¤t->real_parent->children);
340 /* Set the exit signal to SIGCHLD so we signal init on exit */
341 current->exit_signal = SIGCHLD;
343 if (task_nice(current) < 0)
344 set_user_nice(current, 0);
348 security_task_reparent_to_init(current);
349 memcpy(current->signal->rlim, init_task.signal->rlim,
350 sizeof(current->signal->rlim));
351 atomic_inc(&(INIT_USER->__count));
352 write_unlock_irq(&tasklist_lock);
353 switch_uid(INIT_USER);
356 void __set_special_pids(struct pid *pid)
358 struct task_struct *curr = current->group_leader;
359 pid_t nr = pid_nr(pid);
361 if (task_session(curr) != pid) {
362 change_pid(curr, PIDTYPE_SID, pid);
363 set_task_session(curr, nr);
365 if (task_pgrp(curr) != pid) {
366 change_pid(curr, PIDTYPE_PGID, pid);
367 set_task_pgrp(curr, nr);
371 static void set_special_pids(struct pid *pid)
373 write_lock_irq(&tasklist_lock);
374 __set_special_pids(pid);
375 write_unlock_irq(&tasklist_lock);
379 * Let kernel threads use this to say that they
380 * allow a certain signal (since daemonize() will
381 * have disabled all of them by default).
383 int allow_signal(int sig)
385 if (!valid_signal(sig) || sig < 1)
388 spin_lock_irq(¤t->sighand->siglock);
389 sigdelset(¤t->blocked, sig);
391 /* Kernel threads handle their own signals.
392 Let the signal code know it'll be handled, so
393 that they don't get converted to SIGKILL or
394 just silently dropped */
395 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
398 spin_unlock_irq(¤t->sighand->siglock);
402 EXPORT_SYMBOL(allow_signal);
404 int disallow_signal(int sig)
406 if (!valid_signal(sig) || sig < 1)
409 spin_lock_irq(¤t->sighand->siglock);
410 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
412 spin_unlock_irq(¤t->sighand->siglock);
416 EXPORT_SYMBOL(disallow_signal);
419 * Put all the gunge required to become a kernel thread without
420 * attached user resources in one place where it belongs.
423 void daemonize(const char *name, ...)
426 struct fs_struct *fs;
429 va_start(args, name);
430 vsnprintf(current->comm, sizeof(current->comm), name, args);
434 * If we were started as result of loading a module, close all of the
435 * user space pages. We don't need them, and if we didn't close them
436 * they would be locked into memory.
440 * We don't want to have TIF_FREEZE set if the system-wide hibernation
441 * or suspend transition begins right now.
443 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
445 if (current->nsproxy != &init_nsproxy) {
446 get_nsproxy(&init_nsproxy);
447 switch_task_namespaces(current, &init_nsproxy);
449 set_special_pids(&init_struct_pid);
450 proc_clear_tty(current);
452 /* Block and flush all signals */
453 sigfillset(&blocked);
454 sigprocmask(SIG_BLOCK, &blocked, NULL);
455 flush_signals(current);
457 /* Become as one with the init task */
459 exit_fs(current); /* current->fs->count--; */
462 atomic_inc(&fs->count);
465 current->files = init_task.files;
466 atomic_inc(¤t->files->count);
468 reparent_to_kthreadd();
471 EXPORT_SYMBOL(daemonize);
473 static void close_files(struct files_struct * files)
481 * It is safe to dereference the fd table without RCU or
482 * ->file_lock because this is the last reference to the
485 fdt = files_fdtable(files);
489 if (i >= fdt->max_fds)
491 set = fdt->open_fds->fds_bits[j++];
494 struct file * file = xchg(&fdt->fd[i], NULL);
496 filp_close(file, files);
506 struct files_struct *get_files_struct(struct task_struct *task)
508 struct files_struct *files;
513 atomic_inc(&files->count);
519 void put_files_struct(struct files_struct *files)
523 if (atomic_dec_and_test(&files->count)) {
526 * Free the fd and fdset arrays if we expanded them.
527 * If the fdtable was embedded, pass files for freeing
528 * at the end of the RCU grace period. Otherwise,
529 * you can free files immediately.
531 fdt = files_fdtable(files);
532 if (fdt != &files->fdtab)
533 kmem_cache_free(files_cachep, files);
538 void reset_files_struct(struct files_struct *files)
540 struct task_struct *tsk = current;
541 struct files_struct *old;
547 put_files_struct(old);
550 void exit_files(struct task_struct *tsk)
552 struct files_struct * files = tsk->files;
558 put_files_struct(files);
562 void put_fs_struct(struct fs_struct *fs)
564 /* No need to hold fs->lock if we are killing it */
565 if (atomic_dec_and_test(&fs->count)) {
568 kmem_cache_free(fs_cachep, fs);
572 void exit_fs(struct task_struct *tsk)
574 struct fs_struct * fs = tsk->fs;
584 EXPORT_SYMBOL_GPL(exit_fs);
586 #ifdef CONFIG_MM_OWNER
588 * Task p is exiting and it owned mm, lets find a new owner for it
591 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
594 * If there are other users of the mm and the owner (us) is exiting
595 * we need to find a new owner to take on the responsibility.
599 if (atomic_read(&mm->mm_users) <= 1)
606 void mm_update_next_owner(struct mm_struct *mm)
608 struct task_struct *c, *g, *p = current;
611 if (!mm_need_new_owner(mm, p))
614 read_lock(&tasklist_lock);
616 * Search in the children
618 list_for_each_entry(c, &p->children, sibling) {
620 goto assign_new_owner;
624 * Search in the siblings
626 list_for_each_entry(c, &p->parent->children, sibling) {
628 goto assign_new_owner;
632 * Search through everything else. We should not get
635 do_each_thread(g, c) {
637 goto assign_new_owner;
638 } while_each_thread(g, c);
640 read_unlock(&tasklist_lock);
647 * The task_lock protects c->mm from changing.
648 * We always want mm->owner->mm == mm
652 * Delay read_unlock() till we have the task_lock()
653 * to ensure that c does not slip away underneath us
655 read_unlock(&tasklist_lock);
661 cgroup_mm_owner_callbacks(mm->owner, c);
666 #endif /* CONFIG_MM_OWNER */
669 * Turn us into a lazy TLB process if we
672 static void exit_mm(struct task_struct * tsk)
674 struct mm_struct *mm = tsk->mm;
675 struct core_state *core_state;
681 * Serialize with any possible pending coredump.
682 * We must hold mmap_sem around checking core_state
683 * and clearing tsk->mm. The core-inducing thread
684 * will increment ->nr_threads for each thread in the
685 * group with ->mm != NULL.
687 down_read(&mm->mmap_sem);
688 core_state = mm->core_state;
690 struct core_thread self;
691 up_read(&mm->mmap_sem);
694 self.next = xchg(&core_state->dumper.next, &self);
696 * Implies mb(), the result of xchg() must be visible
697 * to core_state->dumper.
699 if (atomic_dec_and_test(&core_state->nr_threads))
700 complete(&core_state->startup);
703 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
704 if (!self.task) /* see coredump_finish() */
708 __set_task_state(tsk, TASK_RUNNING);
709 down_read(&mm->mmap_sem);
711 atomic_inc(&mm->mm_count);
712 BUG_ON(mm != tsk->active_mm);
713 /* more a memory barrier than a real lock */
716 up_read(&mm->mmap_sem);
717 enter_lazy_tlb(mm, current);
718 /* We don't want this task to be frozen prematurely */
719 clear_freeze_flag(tsk);
721 mm_update_next_owner(mm);
726 * Return nonzero if @parent's children should reap themselves.
728 * Called with write_lock_irq(&tasklist_lock) held.
730 static int ignoring_children(struct task_struct *parent)
733 struct sighand_struct *psig = parent->sighand;
735 spin_lock_irqsave(&psig->siglock, flags);
736 ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
737 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
738 spin_unlock_irqrestore(&psig->siglock, flags);
743 * Detach all tasks we were using ptrace on.
744 * Any that need to be release_task'd are put on the @dead list.
746 * Called with write_lock(&tasklist_lock) held.
748 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
750 struct task_struct *p, *n;
753 list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
756 if (p->exit_state != EXIT_ZOMBIE)
760 * If it's a zombie, our attachedness prevented normal
761 * parent notification or self-reaping. Do notification
762 * now if it would have happened earlier. If it should
763 * reap itself, add it to the @dead list. We can't call
764 * release_task() here because we already hold tasklist_lock.
766 * If it's our own child, there is no notification to do.
767 * But if our normal children self-reap, then this child
768 * was prevented by ptrace and we must reap it now.
770 if (!task_detached(p) && thread_group_empty(p)) {
771 if (!same_thread_group(p->real_parent, parent))
772 do_notify_parent(p, p->exit_signal);
775 ign = ignoring_children(parent);
781 if (task_detached(p)) {
783 * Mark it as in the process of being reaped.
785 p->exit_state = EXIT_DEAD;
786 list_add(&p->ptrace_entry, dead);
792 * Finish up exit-time ptrace cleanup.
794 * Called without locks.
796 static void ptrace_exit_finish(struct task_struct *parent,
797 struct list_head *dead)
799 struct task_struct *p, *n;
801 BUG_ON(!list_empty(&parent->ptraced));
803 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
804 list_del_init(&p->ptrace_entry);
809 static void reparent_thread(struct task_struct *p, struct task_struct *father)
811 if (p->pdeath_signal)
812 /* We already hold the tasklist_lock here. */
813 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
815 list_move_tail(&p->sibling, &p->real_parent->children);
817 /* If this is a threaded reparent there is no need to
818 * notify anyone anything has happened.
820 if (same_thread_group(p->real_parent, father))
823 /* We don't want people slaying init. */
824 if (!task_detached(p))
825 p->exit_signal = SIGCHLD;
827 /* If we'd notified the old parent about this child's death,
828 * also notify the new parent.
830 if (!ptrace_reparented(p) &&
831 p->exit_state == EXIT_ZOMBIE &&
832 !task_detached(p) && thread_group_empty(p))
833 do_notify_parent(p, p->exit_signal);
835 kill_orphaned_pgrp(p, father);
839 * When we die, we re-parent all our children.
840 * Try to give them to another thread in our thread
841 * group, and if no such member exists, give it to
842 * the child reaper process (ie "init") in our pid
845 static void forget_original_parent(struct task_struct *father)
847 struct task_struct *p, *n, *reaper = father;
848 LIST_HEAD(ptrace_dead);
850 write_lock_irq(&tasklist_lock);
853 * First clean up ptrace if we were using it.
855 ptrace_exit(father, &ptrace_dead);
858 reaper = next_thread(reaper);
859 if (reaper == father) {
860 reaper = task_child_reaper(father);
863 } while (reaper->flags & PF_EXITING);
865 list_for_each_entry_safe(p, n, &father->children, sibling) {
866 p->real_parent = reaper;
867 if (p->parent == father) {
869 p->parent = p->real_parent;
871 reparent_thread(p, father);
874 write_unlock_irq(&tasklist_lock);
875 BUG_ON(!list_empty(&father->children));
877 ptrace_exit_finish(father, &ptrace_dead);
881 * Send signals to all our closest relatives so that they know
882 * to properly mourn us..
884 static void exit_notify(struct task_struct *tsk, int group_dead)
890 * This does two things:
892 * A. Make init inherit all the child processes
893 * B. Check to see if any process groups have become orphaned
894 * as a result of our exiting, and if they have any stopped
895 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
897 forget_original_parent(tsk);
898 exit_task_namespaces(tsk);
900 write_lock_irq(&tasklist_lock);
902 kill_orphaned_pgrp(tsk->group_leader, NULL);
904 /* Let father know we died
906 * Thread signals are configurable, but you aren't going to use
907 * that to send signals to arbitary processes.
908 * That stops right now.
910 * If the parent exec id doesn't match the exec id we saved
911 * when we started then we know the parent has changed security
914 * If our self_exec id doesn't match our parent_exec_id then
915 * we have changed execution domain as these two values started
916 * the same after a fork.
918 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
919 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
920 tsk->self_exec_id != tsk->parent_exec_id) &&
922 tsk->exit_signal = SIGCHLD;
924 signal = tracehook_notify_death(tsk, &cookie, group_dead);
926 signal = do_notify_parent(tsk, signal);
928 tsk->exit_state = signal < 0 ? EXIT_DEAD : EXIT_ZOMBIE;
930 /* mt-exec, de_thread() is waiting for us */
931 if (thread_group_leader(tsk) &&
932 tsk->signal->notify_count < 0 &&
933 tsk->signal->group_exit_task)
934 wake_up_process(tsk->signal->group_exit_task);
936 write_unlock_irq(&tasklist_lock);
938 tracehook_report_death(tsk, signal, cookie, group_dead);
940 /* If the process is dead, release it - nobody will wait for it */
945 #ifdef CONFIG_DEBUG_STACK_USAGE
946 static void check_stack_usage(void)
948 static DEFINE_SPINLOCK(low_water_lock);
949 static int lowest_to_date = THREAD_SIZE;
950 unsigned long *n = end_of_stack(current);
955 free = (unsigned long)n - (unsigned long)end_of_stack(current);
957 if (free >= lowest_to_date)
960 spin_lock(&low_water_lock);
961 if (free < lowest_to_date) {
962 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
964 current->comm, free);
965 lowest_to_date = free;
967 spin_unlock(&low_water_lock);
970 static inline void check_stack_usage(void) {}
973 static inline void exit_child_reaper(struct task_struct *tsk)
975 if (likely(tsk->group_leader != task_child_reaper(tsk)))
978 if (tsk->nsproxy->pid_ns == &init_pid_ns)
979 panic("Attempted to kill init!");
982 * @tsk is the last thread in the 'cgroup-init' and is exiting.
983 * Terminate all remaining processes in the namespace and reap them
984 * before exiting @tsk.
986 * Note that @tsk (last thread of cgroup-init) may not necessarily
987 * be the child-reaper (i.e main thread of cgroup-init) of the
988 * namespace i.e the child_reaper may have already exited.
990 * Even after a child_reaper exits, we let it inherit orphaned children,
991 * because, pid_ns->child_reaper remains valid as long as there is
992 * at least one living sub-thread in the cgroup init.
994 * This living sub-thread of the cgroup-init will be notified when
995 * a child inherited by the 'child-reaper' exits (do_notify_parent()
996 * uses __group_send_sig_info()). Further, when reaping child processes,
997 * do_wait() iterates over children of all living sub threads.
999 * i.e even though 'child_reaper' thread is listed as the parent of the
1000 * orphaned children, any living sub-thread in the cgroup-init can
1001 * perform the role of the child_reaper.
1003 zap_pid_ns_processes(tsk->nsproxy->pid_ns);
1006 NORET_TYPE void do_exit(long code)
1008 struct task_struct *tsk = current;
1011 profile_task_exit(tsk);
1013 WARN_ON(atomic_read(&tsk->fs_excl));
1015 if (unlikely(in_interrupt()))
1016 panic("Aiee, killing interrupt handler!");
1017 if (unlikely(!tsk->pid))
1018 panic("Attempted to kill the idle task!");
1020 tracehook_report_exit(&code);
1023 * We're taking recursive faults here in do_exit. Safest is to just
1024 * leave this task alone and wait for reboot.
1026 if (unlikely(tsk->flags & PF_EXITING)) {
1028 "Fixing recursive fault but reboot is needed!\n");
1030 * We can do this unlocked here. The futex code uses
1031 * this flag just to verify whether the pi state
1032 * cleanup has been done or not. In the worst case it
1033 * loops once more. We pretend that the cleanup was
1034 * done as there is no way to return. Either the
1035 * OWNER_DIED bit is set by now or we push the blocked
1036 * task into the wait for ever nirwana as well.
1038 tsk->flags |= PF_EXITPIDONE;
1039 if (tsk->io_context)
1041 set_current_state(TASK_UNINTERRUPTIBLE);
1045 exit_signals(tsk); /* sets PF_EXITING */
1047 * tsk->flags are checked in the futex code to protect against
1048 * an exiting task cleaning up the robust pi futexes.
1051 spin_unlock_wait(&tsk->pi_lock);
1053 if (unlikely(in_atomic()))
1054 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1055 current->comm, task_pid_nr(current),
1058 acct_update_integrals(tsk);
1060 update_hiwater_rss(tsk->mm);
1061 update_hiwater_vm(tsk->mm);
1063 group_dead = atomic_dec_and_test(&tsk->signal->live);
1065 exit_child_reaper(tsk);
1066 hrtimer_cancel(&tsk->signal->real_timer);
1067 exit_itimers(tsk->signal);
1069 acct_collect(code, group_dead);
1071 if (unlikely(tsk->robust_list))
1072 exit_robust_list(tsk);
1073 #ifdef CONFIG_COMPAT
1074 if (unlikely(tsk->compat_robust_list))
1075 compat_exit_robust_list(tsk);
1080 if (unlikely(tsk->audit_context))
1083 tsk->exit_code = code;
1084 taskstats_exit(tsk, group_dead);
1093 check_stack_usage();
1095 cgroup_exit(tsk, 1);
1098 if (group_dead && tsk->signal->leader)
1099 disassociate_ctty(1);
1101 module_put(task_thread_info(tsk)->exec_domain->module);
1103 module_put(tsk->binfmt->module);
1105 proc_exit_connector(tsk);
1106 exit_notify(tsk, group_dead);
1108 mpol_put(tsk->mempolicy);
1109 tsk->mempolicy = NULL;
1113 * This must happen late, after the PID is not
1116 if (unlikely(!list_empty(&tsk->pi_state_list)))
1117 exit_pi_state_list(tsk);
1118 if (unlikely(current->pi_state_cache))
1119 kfree(current->pi_state_cache);
1122 * Make sure we are holding no locks:
1124 debug_check_no_locks_held(tsk);
1126 * We can do this unlocked here. The futex code uses this flag
1127 * just to verify whether the pi state cleanup has been done
1128 * or not. In the worst case it loops once more.
1130 tsk->flags |= PF_EXITPIDONE;
1132 if (tsk->io_context)
1135 if (tsk->splice_pipe)
1136 __free_pipe_info(tsk->splice_pipe);
1139 /* causes final put_task_struct in finish_task_switch(). */
1140 tsk->state = TASK_DEAD;
1144 /* Avoid "noreturn function does return". */
1146 cpu_relax(); /* For when BUG is null */
1149 EXPORT_SYMBOL_GPL(do_exit);
1151 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1159 EXPORT_SYMBOL(complete_and_exit);
1161 asmlinkage long sys_exit(int error_code)
1163 do_exit((error_code&0xff)<<8);
1167 * Take down every thread in the group. This is called by fatal signals
1168 * as well as by sys_exit_group (below).
1171 do_group_exit(int exit_code)
1173 struct signal_struct *sig = current->signal;
1175 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1177 if (signal_group_exit(sig))
1178 exit_code = sig->group_exit_code;
1179 else if (!thread_group_empty(current)) {
1180 struct sighand_struct *const sighand = current->sighand;
1181 spin_lock_irq(&sighand->siglock);
1182 if (signal_group_exit(sig))
1183 /* Another thread got here before we took the lock. */
1184 exit_code = sig->group_exit_code;
1186 sig->group_exit_code = exit_code;
1187 sig->flags = SIGNAL_GROUP_EXIT;
1188 zap_other_threads(current);
1190 spin_unlock_irq(&sighand->siglock);
1198 * this kills every thread in the thread group. Note that any externally
1199 * wait4()-ing process will get the correct exit code - even if this
1200 * thread is not the thread group leader.
1202 asmlinkage void sys_exit_group(int error_code)
1204 do_group_exit((error_code & 0xff) << 8);
1207 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1209 struct pid *pid = NULL;
1210 if (type == PIDTYPE_PID)
1211 pid = task->pids[type].pid;
1212 else if (type < PIDTYPE_MAX)
1213 pid = task->group_leader->pids[type].pid;
1217 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1218 struct task_struct *p)
1222 if (type < PIDTYPE_MAX) {
1223 if (task_pid_type(p, type) != pid)
1227 /* Wait for all children (clone and not) if __WALL is set;
1228 * otherwise, wait for clone children *only* if __WCLONE is
1229 * set; otherwise, wait for non-clone children *only*. (Note:
1230 * A "clone" child here is one that reports to its parent
1231 * using a signal other than SIGCHLD.) */
1232 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1233 && !(options & __WALL))
1236 err = security_task_wait(p);
1243 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1244 int why, int status,
1245 struct siginfo __user *infop,
1246 struct rusage __user *rusagep)
1248 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1252 retval = put_user(SIGCHLD, &infop->si_signo);
1254 retval = put_user(0, &infop->si_errno);
1256 retval = put_user((short)why, &infop->si_code);
1258 retval = put_user(pid, &infop->si_pid);
1260 retval = put_user(uid, &infop->si_uid);
1262 retval = put_user(status, &infop->si_status);
1269 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1270 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1271 * the lock and this task is uninteresting. If we return nonzero, we have
1272 * released the lock and the system call should return.
1274 static int wait_task_zombie(struct task_struct *p, int options,
1275 struct siginfo __user *infop,
1276 int __user *stat_addr, struct rusage __user *ru)
1278 unsigned long state;
1279 int retval, status, traced;
1280 pid_t pid = task_pid_vnr(p);
1282 if (!likely(options & WEXITED))
1285 if (unlikely(options & WNOWAIT)) {
1287 int exit_code = p->exit_code;
1291 read_unlock(&tasklist_lock);
1292 if ((exit_code & 0x7f) == 0) {
1294 status = exit_code >> 8;
1296 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1297 status = exit_code & 0x7f;
1299 return wait_noreap_copyout(p, pid, uid, why,
1304 * Try to move the task's state to DEAD
1305 * only one thread is allowed to do this:
1307 state = xchg(&p->exit_state, EXIT_DEAD);
1308 if (state != EXIT_ZOMBIE) {
1309 BUG_ON(state != EXIT_DEAD);
1313 traced = ptrace_reparented(p);
1315 if (likely(!traced)) {
1316 struct signal_struct *psig;
1317 struct signal_struct *sig;
1320 * The resource counters for the group leader are in its
1321 * own task_struct. Those for dead threads in the group
1322 * are in its signal_struct, as are those for the child
1323 * processes it has previously reaped. All these
1324 * accumulate in the parent's signal_struct c* fields.
1326 * We don't bother to take a lock here to protect these
1327 * p->signal fields, because they are only touched by
1328 * __exit_signal, which runs with tasklist_lock
1329 * write-locked anyway, and so is excluded here. We do
1330 * need to protect the access to p->parent->signal fields,
1331 * as other threads in the parent group can be right
1332 * here reaping other children at the same time.
1334 spin_lock_irq(&p->parent->sighand->siglock);
1335 psig = p->parent->signal;
1338 cputime_add(psig->cutime,
1339 cputime_add(p->utime,
1340 cputime_add(sig->utime,
1343 cputime_add(psig->cstime,
1344 cputime_add(p->stime,
1345 cputime_add(sig->stime,
1348 cputime_add(psig->cgtime,
1349 cputime_add(p->gtime,
1350 cputime_add(sig->gtime,
1353 p->min_flt + sig->min_flt + sig->cmin_flt;
1355 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1357 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1359 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1361 task_io_get_inblock(p) +
1362 sig->inblock + sig->cinblock;
1364 task_io_get_oublock(p) +
1365 sig->oublock + sig->coublock;
1366 #ifdef CONFIG_TASK_XACCT
1367 psig->rchar += p->rchar + sig->rchar;
1368 psig->wchar += p->wchar + sig->wchar;
1369 psig->syscr += p->syscr + sig->syscr;
1370 psig->syscw += p->syscw + sig->syscw;
1371 #endif /* CONFIG_TASK_XACCT */
1372 #ifdef CONFIG_TASK_IO_ACCOUNTING
1373 psig->ioac.read_bytes +=
1374 p->ioac.read_bytes + sig->ioac.read_bytes;
1375 psig->ioac.write_bytes +=
1376 p->ioac.write_bytes + sig->ioac.write_bytes;
1377 psig->ioac.cancelled_write_bytes +=
1378 p->ioac.cancelled_write_bytes +
1379 sig->ioac.cancelled_write_bytes;
1380 #endif /* CONFIG_TASK_IO_ACCOUNTING */
1381 spin_unlock_irq(&p->parent->sighand->siglock);
1385 * Now we are sure this task is interesting, and no other
1386 * thread can reap it because we set its state to EXIT_DEAD.
1388 read_unlock(&tasklist_lock);
1390 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1391 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1392 ? p->signal->group_exit_code : p->exit_code;
1393 if (!retval && stat_addr)
1394 retval = put_user(status, stat_addr);
1395 if (!retval && infop)
1396 retval = put_user(SIGCHLD, &infop->si_signo);
1397 if (!retval && infop)
1398 retval = put_user(0, &infop->si_errno);
1399 if (!retval && infop) {
1402 if ((status & 0x7f) == 0) {
1406 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1409 retval = put_user((short)why, &infop->si_code);
1411 retval = put_user(status, &infop->si_status);
1413 if (!retval && infop)
1414 retval = put_user(pid, &infop->si_pid);
1415 if (!retval && infop)
1416 retval = put_user(p->uid, &infop->si_uid);
1421 write_lock_irq(&tasklist_lock);
1422 /* We dropped tasklist, ptracer could die and untrace */
1425 * If this is not a detached task, notify the parent.
1426 * If it's still not detached after that, don't release
1429 if (!task_detached(p)) {
1430 do_notify_parent(p, p->exit_signal);
1431 if (!task_detached(p)) {
1432 p->exit_state = EXIT_ZOMBIE;
1436 write_unlock_irq(&tasklist_lock);
1445 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1446 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1447 * the lock and this task is uninteresting. If we return nonzero, we have
1448 * released the lock and the system call should return.
1450 static int wait_task_stopped(int ptrace, struct task_struct *p,
1451 int options, struct siginfo __user *infop,
1452 int __user *stat_addr, struct rusage __user *ru)
1454 int retval, exit_code, why;
1455 uid_t uid = 0; /* unneeded, required by compiler */
1458 if (!(options & WUNTRACED))
1462 spin_lock_irq(&p->sighand->siglock);
1464 if (unlikely(!task_is_stopped_or_traced(p)))
1467 if (!ptrace && p->signal->group_stop_count > 0)
1469 * A group stop is in progress and this is the group leader.
1470 * We won't report until all threads have stopped.
1474 exit_code = p->exit_code;
1478 if (!unlikely(options & WNOWAIT))
1483 spin_unlock_irq(&p->sighand->siglock);
1488 * Now we are pretty sure this task is interesting.
1489 * Make sure it doesn't get reaped out from under us while we
1490 * give up the lock and then examine it below. We don't want to
1491 * keep holding onto the tasklist_lock while we call getrusage and
1492 * possibly take page faults for user memory.
1495 pid = task_pid_vnr(p);
1496 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1497 read_unlock(&tasklist_lock);
1499 if (unlikely(options & WNOWAIT))
1500 return wait_noreap_copyout(p, pid, uid,
1504 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1505 if (!retval && stat_addr)
1506 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1507 if (!retval && infop)
1508 retval = put_user(SIGCHLD, &infop->si_signo);
1509 if (!retval && infop)
1510 retval = put_user(0, &infop->si_errno);
1511 if (!retval && infop)
1512 retval = put_user((short)why, &infop->si_code);
1513 if (!retval && infop)
1514 retval = put_user(exit_code, &infop->si_status);
1515 if (!retval && infop)
1516 retval = put_user(pid, &infop->si_pid);
1517 if (!retval && infop)
1518 retval = put_user(uid, &infop->si_uid);
1528 * Handle do_wait work for one task in a live, non-stopped state.
1529 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1530 * the lock and this task is uninteresting. If we return nonzero, we have
1531 * released the lock and the system call should return.
1533 static int wait_task_continued(struct task_struct *p, int options,
1534 struct siginfo __user *infop,
1535 int __user *stat_addr, struct rusage __user *ru)
1541 if (!unlikely(options & WCONTINUED))
1544 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1547 spin_lock_irq(&p->sighand->siglock);
1548 /* Re-check with the lock held. */
1549 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1550 spin_unlock_irq(&p->sighand->siglock);
1553 if (!unlikely(options & WNOWAIT))
1554 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1555 spin_unlock_irq(&p->sighand->siglock);
1557 pid = task_pid_vnr(p);
1560 read_unlock(&tasklist_lock);
1563 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1565 if (!retval && stat_addr)
1566 retval = put_user(0xffff, stat_addr);
1570 retval = wait_noreap_copyout(p, pid, uid,
1571 CLD_CONTINUED, SIGCONT,
1573 BUG_ON(retval == 0);
1580 * Consider @p for a wait by @parent.
1582 * -ECHILD should be in *@notask_error before the first call.
1583 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1584 * Returns zero if the search for a child should continue;
1585 * then *@notask_error is 0 if @p is an eligible child,
1586 * or another error from security_task_wait(), or still -ECHILD.
1588 static int wait_consider_task(struct task_struct *parent, int ptrace,
1589 struct task_struct *p, int *notask_error,
1590 enum pid_type type, struct pid *pid, int options,
1591 struct siginfo __user *infop,
1592 int __user *stat_addr, struct rusage __user *ru)
1594 int ret = eligible_child(type, pid, options, p);
1598 if (unlikely(ret < 0)) {
1600 * If we have not yet seen any eligible child,
1601 * then let this error code replace -ECHILD.
1602 * A permission error will give the user a clue
1603 * to look for security policy problems, rather
1604 * than for mysterious wait bugs.
1607 *notask_error = ret;
1610 if (likely(!ptrace) && unlikely(p->ptrace)) {
1612 * This child is hidden by ptrace.
1613 * We aren't allowed to see it now, but eventually we will.
1619 if (p->exit_state == EXIT_DEAD)
1623 * We don't reap group leaders with subthreads.
1625 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1626 return wait_task_zombie(p, options, infop, stat_addr, ru);
1629 * It's stopped or running now, so it might
1630 * later continue, exit, or stop again.
1634 if (task_is_stopped_or_traced(p))
1635 return wait_task_stopped(ptrace, p, options,
1636 infop, stat_addr, ru);
1638 return wait_task_continued(p, options, infop, stat_addr, ru);
1642 * Do the work of do_wait() for one thread in the group, @tsk.
1644 * -ECHILD should be in *@notask_error before the first call.
1645 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1646 * Returns zero if the search for a child should continue; then
1647 * *@notask_error is 0 if there were any eligible children,
1648 * or another error from security_task_wait(), or still -ECHILD.
1650 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1651 enum pid_type type, struct pid *pid, int options,
1652 struct siginfo __user *infop, int __user *stat_addr,
1653 struct rusage __user *ru)
1655 struct task_struct *p;
1657 list_for_each_entry(p, &tsk->children, sibling) {
1659 * Do not consider detached threads.
1661 if (!task_detached(p)) {
1662 int ret = wait_consider_task(tsk, 0, p, notask_error,
1664 infop, stat_addr, ru);
1673 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1674 enum pid_type type, struct pid *pid, int options,
1675 struct siginfo __user *infop, int __user *stat_addr,
1676 struct rusage __user *ru)
1678 struct task_struct *p;
1681 * Traditionally we see ptrace'd stopped tasks regardless of options.
1683 options |= WUNTRACED;
1685 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1686 int ret = wait_consider_task(tsk, 1, p, notask_error,
1688 infop, stat_addr, ru);
1696 static long do_wait(enum pid_type type, struct pid *pid, int options,
1697 struct siginfo __user *infop, int __user *stat_addr,
1698 struct rusage __user *ru)
1700 DECLARE_WAITQUEUE(wait, current);
1701 struct task_struct *tsk;
1704 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1707 * If there is nothing that can match our critiera just get out.
1708 * We will clear @retval to zero if we see any child that might later
1709 * match our criteria, even if we are not able to reap it yet.
1712 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1715 current->state = TASK_INTERRUPTIBLE;
1716 read_lock(&tasklist_lock);
1719 int tsk_result = do_wait_thread(tsk, &retval,
1721 infop, stat_addr, ru);
1723 tsk_result = ptrace_do_wait(tsk, &retval,
1725 infop, stat_addr, ru);
1728 * tasklist_lock is unlocked and we have a final result.
1730 retval = tsk_result;
1734 if (options & __WNOTHREAD)
1736 tsk = next_thread(tsk);
1737 BUG_ON(tsk->signal != current->signal);
1738 } while (tsk != current);
1739 read_unlock(&tasklist_lock);
1741 if (!retval && !(options & WNOHANG)) {
1742 retval = -ERESTARTSYS;
1743 if (!signal_pending(current)) {
1750 current->state = TASK_RUNNING;
1751 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1757 * For a WNOHANG return, clear out all the fields
1758 * we would set so the user can easily tell the
1762 retval = put_user(0, &infop->si_signo);
1764 retval = put_user(0, &infop->si_errno);
1766 retval = put_user(0, &infop->si_code);
1768 retval = put_user(0, &infop->si_pid);
1770 retval = put_user(0, &infop->si_uid);
1772 retval = put_user(0, &infop->si_status);
1778 asmlinkage long sys_waitid(int which, pid_t upid,
1779 struct siginfo __user *infop, int options,
1780 struct rusage __user *ru)
1782 struct pid *pid = NULL;
1786 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1788 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1801 type = PIDTYPE_PGID;
1809 if (type < PIDTYPE_MAX)
1810 pid = find_get_pid(upid);
1811 ret = do_wait(type, pid, options, infop, NULL, ru);
1814 /* avoid REGPARM breakage on x86: */
1815 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1819 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1820 int options, struct rusage __user *ru)
1822 struct pid *pid = NULL;
1826 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1827 __WNOTHREAD|__WCLONE|__WALL))
1832 else if (upid < 0) {
1833 type = PIDTYPE_PGID;
1834 pid = find_get_pid(-upid);
1835 } else if (upid == 0) {
1836 type = PIDTYPE_PGID;
1837 pid = get_pid(task_pgrp(current));
1838 } else /* upid > 0 */ {
1840 pid = find_get_pid(upid);
1843 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1846 /* avoid REGPARM breakage on x86: */
1847 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1851 #ifdef __ARCH_WANT_SYS_WAITPID
1854 * sys_waitpid() remains for compatibility. waitpid() should be
1855 * implemented by calling sys_wait4() from libc.a.
1857 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1859 return sys_wait4(pid, stat_addr, options, NULL);