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));
88 sighand = rcu_dereference(tsk->sighand);
89 spin_lock(&sighand->siglock);
91 posix_cpu_timers_exit(tsk);
92 if (atomic_dec_and_test(&sig->count))
93 posix_cpu_timers_exit_group(tsk);
96 * If there is any task waiting for the group exit
99 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
100 wake_up_process(sig->group_exit_task);
102 if (tsk == sig->curr_target)
103 sig->curr_target = next_thread(tsk);
105 * Accumulate here the counters for all threads but the
106 * group leader as they die, so they can be added into
107 * the process-wide totals when those are taken.
108 * The group leader stays around as a zombie as long
109 * as there are other threads. When it gets reaped,
110 * the exit.c code will add its counts into these totals.
111 * We won't ever get here for the group leader, since it
112 * will have been the last reference on the signal_struct.
114 sig->utime = cputime_add(sig->utime, tsk->utime);
115 sig->stime = cputime_add(sig->stime, tsk->stime);
116 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
117 sig->min_flt += tsk->min_flt;
118 sig->maj_flt += tsk->maj_flt;
119 sig->nvcsw += tsk->nvcsw;
120 sig->nivcsw += tsk->nivcsw;
121 sig->inblock += task_io_get_inblock(tsk);
122 sig->oublock += task_io_get_oublock(tsk);
123 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
124 sig = NULL; /* Marker for below. */
127 __unhash_process(tsk);
130 * Do this under ->siglock, we can race with another thread
131 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
133 flush_sigqueue(&tsk->pending);
137 spin_unlock(&sighand->siglock);
139 __cleanup_sighand(sighand);
140 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
142 flush_sigqueue(&sig->shared_pending);
143 taskstats_tgid_free(sig);
144 __cleanup_signal(sig);
148 static void delayed_put_task_struct(struct rcu_head *rhp)
150 put_task_struct(container_of(rhp, struct task_struct, rcu));
154 * Do final ptrace-related cleanup of a zombie being reaped.
156 * Called with write_lock(&tasklist_lock) held.
158 static void ptrace_release_task(struct task_struct *p)
160 BUG_ON(!list_empty(&p->ptraced));
162 BUG_ON(!list_empty(&p->ptrace_entry));
165 void release_task(struct task_struct * p)
167 struct task_struct *leader;
170 atomic_dec(&p->user->processes);
172 write_lock_irq(&tasklist_lock);
173 ptrace_release_task(p);
177 * If we are the last non-leader member of the thread
178 * group, and the leader is zombie, then notify the
179 * group leader's parent process. (if it wants notification.)
182 leader = p->group_leader;
183 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
184 BUG_ON(task_detached(leader));
185 do_notify_parent(leader, leader->exit_signal);
187 * If we were the last child thread and the leader has
188 * exited already, and the leader's parent ignores SIGCHLD,
189 * then we are the one who should release the leader.
191 * do_notify_parent() will have marked it self-reaping in
194 zap_leader = task_detached(leader);
197 write_unlock_irq(&tasklist_lock);
199 call_rcu(&p->rcu, delayed_put_task_struct);
202 if (unlikely(zap_leader))
207 * This checks not only the pgrp, but falls back on the pid if no
208 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
211 * The caller must hold rcu lock or the tasklist lock.
213 struct pid *session_of_pgrp(struct pid *pgrp)
215 struct task_struct *p;
216 struct pid *sid = NULL;
218 p = pid_task(pgrp, PIDTYPE_PGID);
220 p = pid_task(pgrp, PIDTYPE_PID);
222 sid = task_session(p);
228 * Determine if a process group is "orphaned", according to the POSIX
229 * definition in 2.2.2.52. Orphaned process groups are not to be affected
230 * by terminal-generated stop signals. Newly orphaned process groups are
231 * to receive a SIGHUP and a SIGCONT.
233 * "I ask you, have you ever known what it is to be an orphan?"
235 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
237 struct task_struct *p;
239 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
240 if ((p == ignored_task) ||
241 (p->exit_state && thread_group_empty(p)) ||
242 is_global_init(p->real_parent))
245 if (task_pgrp(p->real_parent) != pgrp &&
246 task_session(p->real_parent) == task_session(p))
248 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
253 int is_current_pgrp_orphaned(void)
257 read_lock(&tasklist_lock);
258 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
259 read_unlock(&tasklist_lock);
264 static int has_stopped_jobs(struct pid *pgrp)
267 struct task_struct *p;
269 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
270 if (!task_is_stopped(p))
274 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
279 * Check to see if any process groups have become orphaned as
280 * a result of our exiting, and if they have any stopped jobs,
281 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
284 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
286 struct pid *pgrp = task_pgrp(tsk);
287 struct task_struct *ignored_task = tsk;
290 /* exit: our father is in a different pgrp than
291 * we are and we were the only connection outside.
293 parent = tsk->real_parent;
295 /* reparent: our child is in a different pgrp than
296 * we are, and it was the only connection outside.
300 if (task_pgrp(parent) != pgrp &&
301 task_session(parent) == task_session(tsk) &&
302 will_become_orphaned_pgrp(pgrp, ignored_task) &&
303 has_stopped_jobs(pgrp)) {
304 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
305 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
310 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
312 * If a kernel thread is launched as a result of a system call, or if
313 * it ever exits, it should generally reparent itself to kthreadd so it
314 * isn't in the way of other processes and is correctly cleaned up on exit.
316 * The various task state such as scheduling policy and priority may have
317 * been inherited from a user process, so we reset them to sane values here.
319 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
321 static void reparent_to_kthreadd(void)
323 write_lock_irq(&tasklist_lock);
325 ptrace_unlink(current);
326 /* Reparent to init */
327 current->real_parent = current->parent = kthreadd_task;
328 list_move_tail(¤t->sibling, ¤t->real_parent->children);
330 /* Set the exit signal to SIGCHLD so we signal init on exit */
331 current->exit_signal = SIGCHLD;
333 if (task_nice(current) < 0)
334 set_user_nice(current, 0);
338 security_task_reparent_to_init(current);
339 memcpy(current->signal->rlim, init_task.signal->rlim,
340 sizeof(current->signal->rlim));
341 atomic_inc(&(INIT_USER->__count));
342 write_unlock_irq(&tasklist_lock);
343 switch_uid(INIT_USER);
346 void __set_special_pids(struct pid *pid)
348 struct task_struct *curr = current->group_leader;
349 pid_t nr = pid_nr(pid);
351 if (task_session(curr) != pid) {
352 change_pid(curr, PIDTYPE_SID, pid);
353 set_task_session(curr, nr);
355 if (task_pgrp(curr) != pid) {
356 change_pid(curr, PIDTYPE_PGID, pid);
357 set_task_pgrp(curr, nr);
361 static void set_special_pids(struct pid *pid)
363 write_lock_irq(&tasklist_lock);
364 __set_special_pids(pid);
365 write_unlock_irq(&tasklist_lock);
369 * Let kernel threads use this to say that they
370 * allow a certain signal (since daemonize() will
371 * have disabled all of them by default).
373 int allow_signal(int sig)
375 if (!valid_signal(sig) || sig < 1)
378 spin_lock_irq(¤t->sighand->siglock);
379 sigdelset(¤t->blocked, sig);
381 /* Kernel threads handle their own signals.
382 Let the signal code know it'll be handled, so
383 that they don't get converted to SIGKILL or
384 just silently dropped */
385 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
388 spin_unlock_irq(¤t->sighand->siglock);
392 EXPORT_SYMBOL(allow_signal);
394 int disallow_signal(int sig)
396 if (!valid_signal(sig) || sig < 1)
399 spin_lock_irq(¤t->sighand->siglock);
400 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
402 spin_unlock_irq(¤t->sighand->siglock);
406 EXPORT_SYMBOL(disallow_signal);
409 * Put all the gunge required to become a kernel thread without
410 * attached user resources in one place where it belongs.
413 void daemonize(const char *name, ...)
416 struct fs_struct *fs;
419 va_start(args, name);
420 vsnprintf(current->comm, sizeof(current->comm), name, args);
424 * If we were started as result of loading a module, close all of the
425 * user space pages. We don't need them, and if we didn't close them
426 * they would be locked into memory.
430 * We don't want to have TIF_FREEZE set if the system-wide hibernation
431 * or suspend transition begins right now.
433 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
435 if (current->nsproxy != &init_nsproxy) {
436 get_nsproxy(&init_nsproxy);
437 switch_task_namespaces(current, &init_nsproxy);
439 set_special_pids(&init_struct_pid);
440 proc_clear_tty(current);
442 /* Block and flush all signals */
443 sigfillset(&blocked);
444 sigprocmask(SIG_BLOCK, &blocked, NULL);
445 flush_signals(current);
447 /* Become as one with the init task */
449 exit_fs(current); /* current->fs->count--; */
452 atomic_inc(&fs->count);
455 current->files = init_task.files;
456 atomic_inc(¤t->files->count);
458 reparent_to_kthreadd();
461 EXPORT_SYMBOL(daemonize);
463 static void close_files(struct files_struct * files)
471 * It is safe to dereference the fd table without RCU or
472 * ->file_lock because this is the last reference to the
475 fdt = files_fdtable(files);
479 if (i >= fdt->max_fds)
481 set = fdt->open_fds->fds_bits[j++];
484 struct file * file = xchg(&fdt->fd[i], NULL);
486 filp_close(file, files);
496 struct files_struct *get_files_struct(struct task_struct *task)
498 struct files_struct *files;
503 atomic_inc(&files->count);
509 void put_files_struct(struct files_struct *files)
513 if (atomic_dec_and_test(&files->count)) {
516 * Free the fd and fdset arrays if we expanded them.
517 * If the fdtable was embedded, pass files for freeing
518 * at the end of the RCU grace period. Otherwise,
519 * you can free files immediately.
521 fdt = files_fdtable(files);
522 if (fdt != &files->fdtab)
523 kmem_cache_free(files_cachep, files);
528 void reset_files_struct(struct files_struct *files)
530 struct task_struct *tsk = current;
531 struct files_struct *old;
537 put_files_struct(old);
540 void exit_files(struct task_struct *tsk)
542 struct files_struct * files = tsk->files;
548 put_files_struct(files);
552 void put_fs_struct(struct fs_struct *fs)
554 /* No need to hold fs->lock if we are killing it */
555 if (atomic_dec_and_test(&fs->count)) {
558 if (fs->altroot.dentry)
559 path_put(&fs->altroot);
560 kmem_cache_free(fs_cachep, fs);
564 void exit_fs(struct task_struct *tsk)
566 struct fs_struct * fs = tsk->fs;
576 EXPORT_SYMBOL_GPL(exit_fs);
578 #ifdef CONFIG_MM_OWNER
580 * Task p is exiting and it owned mm, lets find a new owner for it
583 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
586 * If there are other users of the mm and the owner (us) is exiting
587 * we need to find a new owner to take on the responsibility.
591 if (atomic_read(&mm->mm_users) <= 1)
598 void mm_update_next_owner(struct mm_struct *mm)
600 struct task_struct *c, *g, *p = current;
603 if (!mm_need_new_owner(mm, p))
606 read_lock(&tasklist_lock);
608 * Search in the children
610 list_for_each_entry(c, &p->children, sibling) {
612 goto assign_new_owner;
616 * Search in the siblings
618 list_for_each_entry(c, &p->parent->children, sibling) {
620 goto assign_new_owner;
624 * Search through everything else. We should not get
627 do_each_thread(g, c) {
629 goto assign_new_owner;
630 } while_each_thread(g, c);
632 read_unlock(&tasklist_lock);
639 * The task_lock protects c->mm from changing.
640 * We always want mm->owner->mm == mm
644 * Delay read_unlock() till we have the task_lock()
645 * to ensure that c does not slip away underneath us
647 read_unlock(&tasklist_lock);
653 cgroup_mm_owner_callbacks(mm->owner, c);
658 #endif /* CONFIG_MM_OWNER */
661 * Turn us into a lazy TLB process if we
664 static void exit_mm(struct task_struct * tsk)
666 struct mm_struct *mm = tsk->mm;
667 struct core_state *core_state;
673 * Serialize with any possible pending coredump.
674 * We must hold mmap_sem around checking core_state
675 * and clearing tsk->mm. The core-inducing thread
676 * will increment ->nr_threads for each thread in the
677 * group with ->mm != NULL.
679 down_read(&mm->mmap_sem);
680 core_state = mm->core_state;
682 struct core_thread self;
683 up_read(&mm->mmap_sem);
686 self.next = xchg(&core_state->dumper.next, &self);
688 * Implies mb(), the result of xchg() must be visible
689 * to core_state->dumper.
691 if (atomic_dec_and_test(&core_state->nr_threads))
692 complete(&core_state->startup);
694 wait_for_completion(&mm->core_done);
695 down_read(&mm->mmap_sem);
697 atomic_inc(&mm->mm_count);
698 BUG_ON(mm != tsk->active_mm);
699 /* more a memory barrier than a real lock */
702 up_read(&mm->mmap_sem);
703 enter_lazy_tlb(mm, current);
704 /* We don't want this task to be frozen prematurely */
705 clear_freeze_flag(tsk);
707 mm_update_next_owner(mm);
712 * Return nonzero if @parent's children should reap themselves.
714 * Called with write_lock_irq(&tasklist_lock) held.
716 static int ignoring_children(struct task_struct *parent)
719 struct sighand_struct *psig = parent->sighand;
721 spin_lock_irqsave(&psig->siglock, flags);
722 ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
723 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
724 spin_unlock_irqrestore(&psig->siglock, flags);
729 * Detach all tasks we were using ptrace on.
730 * Any that need to be release_task'd are put on the @dead list.
732 * Called with write_lock(&tasklist_lock) held.
734 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
736 struct task_struct *p, *n;
739 list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
742 if (p->exit_state != EXIT_ZOMBIE)
746 * If it's a zombie, our attachedness prevented normal
747 * parent notification or self-reaping. Do notification
748 * now if it would have happened earlier. If it should
749 * reap itself, add it to the @dead list. We can't call
750 * release_task() here because we already hold tasklist_lock.
752 * If it's our own child, there is no notification to do.
753 * But if our normal children self-reap, then this child
754 * was prevented by ptrace and we must reap it now.
756 if (!task_detached(p) && thread_group_empty(p)) {
757 if (!same_thread_group(p->real_parent, parent))
758 do_notify_parent(p, p->exit_signal);
761 ign = ignoring_children(parent);
767 if (task_detached(p)) {
769 * Mark it as in the process of being reaped.
771 p->exit_state = EXIT_DEAD;
772 list_add(&p->ptrace_entry, dead);
778 * Finish up exit-time ptrace cleanup.
780 * Called without locks.
782 static void ptrace_exit_finish(struct task_struct *parent,
783 struct list_head *dead)
785 struct task_struct *p, *n;
787 BUG_ON(!list_empty(&parent->ptraced));
789 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
790 list_del_init(&p->ptrace_entry);
795 static void reparent_thread(struct task_struct *p, struct task_struct *father)
797 if (p->pdeath_signal)
798 /* We already hold the tasklist_lock here. */
799 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
801 list_move_tail(&p->sibling, &p->real_parent->children);
803 /* If this is a threaded reparent there is no need to
804 * notify anyone anything has happened.
806 if (same_thread_group(p->real_parent, father))
809 /* We don't want people slaying init. */
810 if (!task_detached(p))
811 p->exit_signal = SIGCHLD;
813 /* If we'd notified the old parent about this child's death,
814 * also notify the new parent.
816 if (!ptrace_reparented(p) &&
817 p->exit_state == EXIT_ZOMBIE &&
818 !task_detached(p) && thread_group_empty(p))
819 do_notify_parent(p, p->exit_signal);
821 kill_orphaned_pgrp(p, father);
825 * When we die, we re-parent all our children.
826 * Try to give them to another thread in our thread
827 * group, and if no such member exists, give it to
828 * the child reaper process (ie "init") in our pid
831 static void forget_original_parent(struct task_struct *father)
833 struct task_struct *p, *n, *reaper = father;
834 LIST_HEAD(ptrace_dead);
836 write_lock_irq(&tasklist_lock);
839 * First clean up ptrace if we were using it.
841 ptrace_exit(father, &ptrace_dead);
844 reaper = next_thread(reaper);
845 if (reaper == father) {
846 reaper = task_child_reaper(father);
849 } while (reaper->flags & PF_EXITING);
851 list_for_each_entry_safe(p, n, &father->children, sibling) {
852 p->real_parent = reaper;
853 if (p->parent == father) {
855 p->parent = p->real_parent;
857 reparent_thread(p, father);
860 write_unlock_irq(&tasklist_lock);
861 BUG_ON(!list_empty(&father->children));
863 ptrace_exit_finish(father, &ptrace_dead);
867 * Send signals to all our closest relatives so that they know
868 * to properly mourn us..
870 static void exit_notify(struct task_struct *tsk, int group_dead)
875 * This does two things:
877 * A. Make init inherit all the child processes
878 * B. Check to see if any process groups have become orphaned
879 * as a result of our exiting, and if they have any stopped
880 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
882 forget_original_parent(tsk);
883 exit_task_namespaces(tsk);
885 write_lock_irq(&tasklist_lock);
887 kill_orphaned_pgrp(tsk->group_leader, NULL);
889 /* Let father know we died
891 * Thread signals are configurable, but you aren't going to use
892 * that to send signals to arbitary processes.
893 * That stops right now.
895 * If the parent exec id doesn't match the exec id we saved
896 * when we started then we know the parent has changed security
899 * If our self_exec id doesn't match our parent_exec_id then
900 * we have changed execution domain as these two values started
901 * the same after a fork.
903 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
904 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
905 tsk->self_exec_id != tsk->parent_exec_id) &&
907 tsk->exit_signal = SIGCHLD;
909 /* If something other than our normal parent is ptracing us, then
910 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
911 * only has special meaning to our real parent.
913 if (!task_detached(tsk) && thread_group_empty(tsk)) {
914 int signal = ptrace_reparented(tsk) ?
915 SIGCHLD : tsk->exit_signal;
916 do_notify_parent(tsk, signal);
917 } else if (tsk->ptrace) {
918 do_notify_parent(tsk, SIGCHLD);
922 if (task_detached(tsk) && likely(!tsk->ptrace))
924 tsk->exit_state = state;
926 /* mt-exec, de_thread() is waiting for us */
927 if (thread_group_leader(tsk) &&
928 tsk->signal->notify_count < 0 &&
929 tsk->signal->group_exit_task)
930 wake_up_process(tsk->signal->group_exit_task);
932 write_unlock_irq(&tasklist_lock);
934 /* If the process is dead, release it - nobody will wait for it */
935 if (state == EXIT_DEAD)
939 #ifdef CONFIG_DEBUG_STACK_USAGE
940 static void check_stack_usage(void)
942 static DEFINE_SPINLOCK(low_water_lock);
943 static int lowest_to_date = THREAD_SIZE;
944 unsigned long *n = end_of_stack(current);
949 free = (unsigned long)n - (unsigned long)end_of_stack(current);
951 if (free >= lowest_to_date)
954 spin_lock(&low_water_lock);
955 if (free < lowest_to_date) {
956 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
958 current->comm, free);
959 lowest_to_date = free;
961 spin_unlock(&low_water_lock);
964 static inline void check_stack_usage(void) {}
967 static inline void exit_child_reaper(struct task_struct *tsk)
969 if (likely(tsk->group_leader != task_child_reaper(tsk)))
972 if (tsk->nsproxy->pid_ns == &init_pid_ns)
973 panic("Attempted to kill init!");
976 * @tsk is the last thread in the 'cgroup-init' and is exiting.
977 * Terminate all remaining processes in the namespace and reap them
978 * before exiting @tsk.
980 * Note that @tsk (last thread of cgroup-init) may not necessarily
981 * be the child-reaper (i.e main thread of cgroup-init) of the
982 * namespace i.e the child_reaper may have already exited.
984 * Even after a child_reaper exits, we let it inherit orphaned children,
985 * because, pid_ns->child_reaper remains valid as long as there is
986 * at least one living sub-thread in the cgroup init.
988 * This living sub-thread of the cgroup-init will be notified when
989 * a child inherited by the 'child-reaper' exits (do_notify_parent()
990 * uses __group_send_sig_info()). Further, when reaping child processes,
991 * do_wait() iterates over children of all living sub threads.
993 * i.e even though 'child_reaper' thread is listed as the parent of the
994 * orphaned children, any living sub-thread in the cgroup-init can
995 * perform the role of the child_reaper.
997 zap_pid_ns_processes(tsk->nsproxy->pid_ns);
1000 NORET_TYPE void do_exit(long code)
1002 struct task_struct *tsk = current;
1005 profile_task_exit(tsk);
1007 WARN_ON(atomic_read(&tsk->fs_excl));
1009 if (unlikely(in_interrupt()))
1010 panic("Aiee, killing interrupt handler!");
1011 if (unlikely(!tsk->pid))
1012 panic("Attempted to kill the idle task!");
1014 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
1015 current->ptrace_message = code;
1016 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
1020 * We're taking recursive faults here in do_exit. Safest is to just
1021 * leave this task alone and wait for reboot.
1023 if (unlikely(tsk->flags & PF_EXITING)) {
1025 "Fixing recursive fault but reboot is needed!\n");
1027 * We can do this unlocked here. The futex code uses
1028 * this flag just to verify whether the pi state
1029 * cleanup has been done or not. In the worst case it
1030 * loops once more. We pretend that the cleanup was
1031 * done as there is no way to return. Either the
1032 * OWNER_DIED bit is set by now or we push the blocked
1033 * task into the wait for ever nirwana as well.
1035 tsk->flags |= PF_EXITPIDONE;
1036 if (tsk->io_context)
1038 set_current_state(TASK_UNINTERRUPTIBLE);
1042 exit_signals(tsk); /* sets PF_EXITING */
1044 * tsk->flags are checked in the futex code to protect against
1045 * an exiting task cleaning up the robust pi futexes.
1048 spin_unlock_wait(&tsk->pi_lock);
1050 if (unlikely(in_atomic()))
1051 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1052 current->comm, task_pid_nr(current),
1055 acct_update_integrals(tsk);
1057 update_hiwater_rss(tsk->mm);
1058 update_hiwater_vm(tsk->mm);
1060 group_dead = atomic_dec_and_test(&tsk->signal->live);
1062 exit_child_reaper(tsk);
1063 hrtimer_cancel(&tsk->signal->real_timer);
1064 exit_itimers(tsk->signal);
1066 acct_collect(code, group_dead);
1068 if (unlikely(tsk->robust_list))
1069 exit_robust_list(tsk);
1070 #ifdef CONFIG_COMPAT
1071 if (unlikely(tsk->compat_robust_list))
1072 compat_exit_robust_list(tsk);
1077 if (unlikely(tsk->audit_context))
1080 tsk->exit_code = code;
1081 taskstats_exit(tsk, group_dead);
1090 check_stack_usage();
1092 cgroup_exit(tsk, 1);
1095 if (group_dead && tsk->signal->leader)
1096 disassociate_ctty(1);
1098 module_put(task_thread_info(tsk)->exec_domain->module);
1100 module_put(tsk->binfmt->module);
1102 proc_exit_connector(tsk);
1103 exit_notify(tsk, group_dead);
1105 mpol_put(tsk->mempolicy);
1106 tsk->mempolicy = NULL;
1110 * This must happen late, after the PID is not
1113 if (unlikely(!list_empty(&tsk->pi_state_list)))
1114 exit_pi_state_list(tsk);
1115 if (unlikely(current->pi_state_cache))
1116 kfree(current->pi_state_cache);
1119 * Make sure we are holding no locks:
1121 debug_check_no_locks_held(tsk);
1123 * We can do this unlocked here. The futex code uses this flag
1124 * just to verify whether the pi state cleanup has been done
1125 * or not. In the worst case it loops once more.
1127 tsk->flags |= PF_EXITPIDONE;
1129 if (tsk->io_context)
1132 if (tsk->splice_pipe)
1133 __free_pipe_info(tsk->splice_pipe);
1136 /* causes final put_task_struct in finish_task_switch(). */
1137 tsk->state = TASK_DEAD;
1141 /* Avoid "noreturn function does return". */
1143 cpu_relax(); /* For when BUG is null */
1146 EXPORT_SYMBOL_GPL(do_exit);
1148 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1156 EXPORT_SYMBOL(complete_and_exit);
1158 asmlinkage long sys_exit(int error_code)
1160 do_exit((error_code&0xff)<<8);
1164 * Take down every thread in the group. This is called by fatal signals
1165 * as well as by sys_exit_group (below).
1168 do_group_exit(int exit_code)
1170 struct signal_struct *sig = current->signal;
1172 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1174 if (signal_group_exit(sig))
1175 exit_code = sig->group_exit_code;
1176 else if (!thread_group_empty(current)) {
1177 struct sighand_struct *const sighand = current->sighand;
1178 spin_lock_irq(&sighand->siglock);
1179 if (signal_group_exit(sig))
1180 /* Another thread got here before we took the lock. */
1181 exit_code = sig->group_exit_code;
1183 sig->group_exit_code = exit_code;
1184 sig->flags = SIGNAL_GROUP_EXIT;
1185 zap_other_threads(current);
1187 spin_unlock_irq(&sighand->siglock);
1195 * this kills every thread in the thread group. Note that any externally
1196 * wait4()-ing process will get the correct exit code - even if this
1197 * thread is not the thread group leader.
1199 asmlinkage void sys_exit_group(int error_code)
1201 do_group_exit((error_code & 0xff) << 8);
1204 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1206 struct pid *pid = NULL;
1207 if (type == PIDTYPE_PID)
1208 pid = task->pids[type].pid;
1209 else if (type < PIDTYPE_MAX)
1210 pid = task->group_leader->pids[type].pid;
1214 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1215 struct task_struct *p)
1219 if (type < PIDTYPE_MAX) {
1220 if (task_pid_type(p, type) != pid)
1224 /* Wait for all children (clone and not) if __WALL is set;
1225 * otherwise, wait for clone children *only* if __WCLONE is
1226 * set; otherwise, wait for non-clone children *only*. (Note:
1227 * A "clone" child here is one that reports to its parent
1228 * using a signal other than SIGCHLD.) */
1229 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1230 && !(options & __WALL))
1233 err = security_task_wait(p);
1240 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1241 int why, int status,
1242 struct siginfo __user *infop,
1243 struct rusage __user *rusagep)
1245 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1249 retval = put_user(SIGCHLD, &infop->si_signo);
1251 retval = put_user(0, &infop->si_errno);
1253 retval = put_user((short)why, &infop->si_code);
1255 retval = put_user(pid, &infop->si_pid);
1257 retval = put_user(uid, &infop->si_uid);
1259 retval = put_user(status, &infop->si_status);
1266 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1267 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1268 * the lock and this task is uninteresting. If we return nonzero, we have
1269 * released the lock and the system call should return.
1271 static int wait_task_zombie(struct task_struct *p, int options,
1272 struct siginfo __user *infop,
1273 int __user *stat_addr, struct rusage __user *ru)
1275 unsigned long state;
1276 int retval, status, traced;
1277 pid_t pid = task_pid_vnr(p);
1279 if (!likely(options & WEXITED))
1282 if (unlikely(options & WNOWAIT)) {
1284 int exit_code = p->exit_code;
1288 read_unlock(&tasklist_lock);
1289 if ((exit_code & 0x7f) == 0) {
1291 status = exit_code >> 8;
1293 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1294 status = exit_code & 0x7f;
1296 return wait_noreap_copyout(p, pid, uid, why,
1301 * Try to move the task's state to DEAD
1302 * only one thread is allowed to do this:
1304 state = xchg(&p->exit_state, EXIT_DEAD);
1305 if (state != EXIT_ZOMBIE) {
1306 BUG_ON(state != EXIT_DEAD);
1310 traced = ptrace_reparented(p);
1312 if (likely(!traced)) {
1313 struct signal_struct *psig;
1314 struct signal_struct *sig;
1317 * The resource counters for the group leader are in its
1318 * own task_struct. Those for dead threads in the group
1319 * are in its signal_struct, as are those for the child
1320 * processes it has previously reaped. All these
1321 * accumulate in the parent's signal_struct c* fields.
1323 * We don't bother to take a lock here to protect these
1324 * p->signal fields, because they are only touched by
1325 * __exit_signal, which runs with tasklist_lock
1326 * write-locked anyway, and so is excluded here. We do
1327 * need to protect the access to p->parent->signal fields,
1328 * as other threads in the parent group can be right
1329 * here reaping other children at the same time.
1331 spin_lock_irq(&p->parent->sighand->siglock);
1332 psig = p->parent->signal;
1335 cputime_add(psig->cutime,
1336 cputime_add(p->utime,
1337 cputime_add(sig->utime,
1340 cputime_add(psig->cstime,
1341 cputime_add(p->stime,
1342 cputime_add(sig->stime,
1345 cputime_add(psig->cgtime,
1346 cputime_add(p->gtime,
1347 cputime_add(sig->gtime,
1350 p->min_flt + sig->min_flt + sig->cmin_flt;
1352 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1354 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1356 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1358 task_io_get_inblock(p) +
1359 sig->inblock + sig->cinblock;
1361 task_io_get_oublock(p) +
1362 sig->oublock + sig->coublock;
1363 spin_unlock_irq(&p->parent->sighand->siglock);
1367 * Now we are sure this task is interesting, and no other
1368 * thread can reap it because we set its state to EXIT_DEAD.
1370 read_unlock(&tasklist_lock);
1372 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1373 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1374 ? p->signal->group_exit_code : p->exit_code;
1375 if (!retval && stat_addr)
1376 retval = put_user(status, stat_addr);
1377 if (!retval && infop)
1378 retval = put_user(SIGCHLD, &infop->si_signo);
1379 if (!retval && infop)
1380 retval = put_user(0, &infop->si_errno);
1381 if (!retval && infop) {
1384 if ((status & 0x7f) == 0) {
1388 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1391 retval = put_user((short)why, &infop->si_code);
1393 retval = put_user(status, &infop->si_status);
1395 if (!retval && infop)
1396 retval = put_user(pid, &infop->si_pid);
1397 if (!retval && infop)
1398 retval = put_user(p->uid, &infop->si_uid);
1403 write_lock_irq(&tasklist_lock);
1404 /* We dropped tasklist, ptracer could die and untrace */
1407 * If this is not a detached task, notify the parent.
1408 * If it's still not detached after that, don't release
1411 if (!task_detached(p)) {
1412 do_notify_parent(p, p->exit_signal);
1413 if (!task_detached(p)) {
1414 p->exit_state = EXIT_ZOMBIE;
1418 write_unlock_irq(&tasklist_lock);
1427 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1428 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1429 * the lock and this task is uninteresting. If we return nonzero, we have
1430 * released the lock and the system call should return.
1432 static int wait_task_stopped(int ptrace, struct task_struct *p,
1433 int options, struct siginfo __user *infop,
1434 int __user *stat_addr, struct rusage __user *ru)
1436 int retval, exit_code, why;
1437 uid_t uid = 0; /* unneeded, required by compiler */
1440 if (!(options & WUNTRACED))
1444 spin_lock_irq(&p->sighand->siglock);
1446 if (unlikely(!task_is_stopped_or_traced(p)))
1449 if (!ptrace && p->signal->group_stop_count > 0)
1451 * A group stop is in progress and this is the group leader.
1452 * We won't report until all threads have stopped.
1456 exit_code = p->exit_code;
1460 if (!unlikely(options & WNOWAIT))
1465 spin_unlock_irq(&p->sighand->siglock);
1470 * Now we are pretty sure this task is interesting.
1471 * Make sure it doesn't get reaped out from under us while we
1472 * give up the lock and then examine it below. We don't want to
1473 * keep holding onto the tasklist_lock while we call getrusage and
1474 * possibly take page faults for user memory.
1477 pid = task_pid_vnr(p);
1478 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1479 read_unlock(&tasklist_lock);
1481 if (unlikely(options & WNOWAIT))
1482 return wait_noreap_copyout(p, pid, uid,
1486 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1487 if (!retval && stat_addr)
1488 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1489 if (!retval && infop)
1490 retval = put_user(SIGCHLD, &infop->si_signo);
1491 if (!retval && infop)
1492 retval = put_user(0, &infop->si_errno);
1493 if (!retval && infop)
1494 retval = put_user((short)why, &infop->si_code);
1495 if (!retval && infop)
1496 retval = put_user(exit_code, &infop->si_status);
1497 if (!retval && infop)
1498 retval = put_user(pid, &infop->si_pid);
1499 if (!retval && infop)
1500 retval = put_user(uid, &infop->si_uid);
1510 * Handle do_wait work for one task in a live, non-stopped state.
1511 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1512 * the lock and this task is uninteresting. If we return nonzero, we have
1513 * released the lock and the system call should return.
1515 static int wait_task_continued(struct task_struct *p, int options,
1516 struct siginfo __user *infop,
1517 int __user *stat_addr, struct rusage __user *ru)
1523 if (!unlikely(options & WCONTINUED))
1526 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1529 spin_lock_irq(&p->sighand->siglock);
1530 /* Re-check with the lock held. */
1531 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1532 spin_unlock_irq(&p->sighand->siglock);
1535 if (!unlikely(options & WNOWAIT))
1536 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1537 spin_unlock_irq(&p->sighand->siglock);
1539 pid = task_pid_vnr(p);
1542 read_unlock(&tasklist_lock);
1545 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1547 if (!retval && stat_addr)
1548 retval = put_user(0xffff, stat_addr);
1552 retval = wait_noreap_copyout(p, pid, uid,
1553 CLD_CONTINUED, SIGCONT,
1555 BUG_ON(retval == 0);
1562 * Consider @p for a wait by @parent.
1564 * -ECHILD should be in *@notask_error before the first call.
1565 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1566 * Returns zero if the search for a child should continue;
1567 * then *@notask_error is 0 if @p is an eligible child,
1568 * or another error from security_task_wait(), or still -ECHILD.
1570 static int wait_consider_task(struct task_struct *parent, int ptrace,
1571 struct task_struct *p, int *notask_error,
1572 enum pid_type type, struct pid *pid, int options,
1573 struct siginfo __user *infop,
1574 int __user *stat_addr, struct rusage __user *ru)
1576 int ret = eligible_child(type, pid, options, p);
1580 if (unlikely(ret < 0)) {
1582 * If we have not yet seen any eligible child,
1583 * then let this error code replace -ECHILD.
1584 * A permission error will give the user a clue
1585 * to look for security policy problems, rather
1586 * than for mysterious wait bugs.
1589 *notask_error = ret;
1592 if (likely(!ptrace) && unlikely(p->ptrace)) {
1594 * This child is hidden by ptrace.
1595 * We aren't allowed to see it now, but eventually we will.
1601 if (p->exit_state == EXIT_DEAD)
1605 * We don't reap group leaders with subthreads.
1607 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1608 return wait_task_zombie(p, options, infop, stat_addr, ru);
1611 * It's stopped or running now, so it might
1612 * later continue, exit, or stop again.
1616 if (task_is_stopped_or_traced(p))
1617 return wait_task_stopped(ptrace, p, options,
1618 infop, stat_addr, ru);
1620 return wait_task_continued(p, options, infop, stat_addr, ru);
1624 * Do the work of do_wait() for one thread in the group, @tsk.
1626 * -ECHILD should be in *@notask_error before the first call.
1627 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1628 * Returns zero if the search for a child should continue; then
1629 * *@notask_error is 0 if there were any eligible children,
1630 * or another error from security_task_wait(), or still -ECHILD.
1632 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1633 enum pid_type type, struct pid *pid, int options,
1634 struct siginfo __user *infop, int __user *stat_addr,
1635 struct rusage __user *ru)
1637 struct task_struct *p;
1639 list_for_each_entry(p, &tsk->children, sibling) {
1641 * Do not consider detached threads.
1643 if (!task_detached(p)) {
1644 int ret = wait_consider_task(tsk, 0, p, notask_error,
1646 infop, stat_addr, ru);
1655 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1656 enum pid_type type, struct pid *pid, int options,
1657 struct siginfo __user *infop, int __user *stat_addr,
1658 struct rusage __user *ru)
1660 struct task_struct *p;
1663 * Traditionally we see ptrace'd stopped tasks regardless of options.
1665 options |= WUNTRACED;
1667 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1668 int ret = wait_consider_task(tsk, 1, p, notask_error,
1670 infop, stat_addr, ru);
1678 static long do_wait(enum pid_type type, struct pid *pid, int options,
1679 struct siginfo __user *infop, int __user *stat_addr,
1680 struct rusage __user *ru)
1682 DECLARE_WAITQUEUE(wait, current);
1683 struct task_struct *tsk;
1686 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1689 * If there is nothing that can match our critiera just get out.
1690 * We will clear @retval to zero if we see any child that might later
1691 * match our criteria, even if we are not able to reap it yet.
1694 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1697 current->state = TASK_INTERRUPTIBLE;
1698 read_lock(&tasklist_lock);
1701 int tsk_result = do_wait_thread(tsk, &retval,
1703 infop, stat_addr, ru);
1705 tsk_result = ptrace_do_wait(tsk, &retval,
1707 infop, stat_addr, ru);
1710 * tasklist_lock is unlocked and we have a final result.
1712 retval = tsk_result;
1716 if (options & __WNOTHREAD)
1718 tsk = next_thread(tsk);
1719 BUG_ON(tsk->signal != current->signal);
1720 } while (tsk != current);
1721 read_unlock(&tasklist_lock);
1723 if (!retval && !(options & WNOHANG)) {
1724 retval = -ERESTARTSYS;
1725 if (!signal_pending(current)) {
1732 current->state = TASK_RUNNING;
1733 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1739 * For a WNOHANG return, clear out all the fields
1740 * we would set so the user can easily tell the
1744 retval = put_user(0, &infop->si_signo);
1746 retval = put_user(0, &infop->si_errno);
1748 retval = put_user(0, &infop->si_code);
1750 retval = put_user(0, &infop->si_pid);
1752 retval = put_user(0, &infop->si_uid);
1754 retval = put_user(0, &infop->si_status);
1760 asmlinkage long sys_waitid(int which, pid_t upid,
1761 struct siginfo __user *infop, int options,
1762 struct rusage __user *ru)
1764 struct pid *pid = NULL;
1768 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1770 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1783 type = PIDTYPE_PGID;
1791 if (type < PIDTYPE_MAX)
1792 pid = find_get_pid(upid);
1793 ret = do_wait(type, pid, options, infop, NULL, ru);
1796 /* avoid REGPARM breakage on x86: */
1797 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1801 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1802 int options, struct rusage __user *ru)
1804 struct pid *pid = NULL;
1808 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1809 __WNOTHREAD|__WCLONE|__WALL))
1814 else if (upid < 0) {
1815 type = PIDTYPE_PGID;
1816 pid = find_get_pid(-upid);
1817 } else if (upid == 0) {
1818 type = PIDTYPE_PGID;
1819 pid = get_pid(task_pgrp(current));
1820 } else /* upid > 0 */ {
1822 pid = find_get_pid(upid);
1825 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1828 /* avoid REGPARM breakage on x86: */
1829 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1833 #ifdef __ARCH_WANT_SYS_WAITPID
1836 * sys_waitpid() remains for compatibility. waitpid() should be
1837 * implemented by calling sys_wait4() from libc.a.
1839 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1841 return sys_wait4(pid, stat_addr, options, NULL);