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/iocontext.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/fdtable.h>
23 #include <linux/binfmts.h>
24 #include <linux/nsproxy.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/ptrace.h>
27 #include <linux/profile.h>
28 #include <linux/mount.h>
29 #include <linux/proc_fs.h>
30 #include <linux/kthread.h>
31 #include <linux/mempolicy.h>
32 #include <linux/taskstats_kern.h>
33 #include <linux/delayacct.h>
34 #include <linux/freezer.h>
35 #include <linux/cgroup.h>
36 #include <linux/syscalls.h>
37 #include <linux/signal.h>
38 #include <linux/posix-timers.h>
39 #include <linux/cn_proc.h>
40 #include <linux/mutex.h>
41 #include <linux/futex.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/tracehook.h>
48 #include <linux/fs_struct.h>
49 #include <linux/init_task.h>
50 #include <linux/perf_event.h>
51 #include <trace/events/sched.h>
52 #include <linux/hw_breakpoint.h>
54 #include <asm/uaccess.h>
55 #include <asm/unistd.h>
56 #include <asm/pgtable.h>
57 #include <asm/mmu_context.h>
59 static void exit_mm(struct task_struct * tsk);
61 static void __unhash_process(struct task_struct *p, bool group_dead)
64 detach_pid(p, PIDTYPE_PID);
66 detach_pid(p, PIDTYPE_PGID);
67 detach_pid(p, PIDTYPE_SID);
69 list_del_rcu(&p->tasks);
70 list_del_init(&p->sibling);
71 __get_cpu_var(process_counts)--;
73 list_del_rcu(&p->thread_group);
77 * This function expects the tasklist_lock write-locked.
79 static void __exit_signal(struct task_struct *tsk)
81 struct signal_struct *sig = tsk->signal;
82 bool group_dead = thread_group_leader(tsk);
83 struct sighand_struct *sighand;
84 struct tty_struct *uninitialized_var(tty);
86 sighand = rcu_dereference_check(tsk->sighand,
87 rcu_read_lock_held() ||
88 lockdep_tasklist_lock_is_held());
89 spin_lock(&sighand->siglock);
91 posix_cpu_timers_exit(tsk);
93 posix_cpu_timers_exit_group(tsk);
98 * If there is any task waiting for the group exit
101 if (sig->notify_count > 0 && !--sig->notify_count)
102 wake_up_process(sig->group_exit_task);
104 if (tsk == sig->curr_target)
105 sig->curr_target = next_thread(tsk);
107 * Accumulate here the counters for all threads but the
108 * group leader as they die, so they can be added into
109 * the process-wide totals when those are taken.
110 * The group leader stays around as a zombie as long
111 * as there are other threads. When it gets reaped,
112 * the exit.c code will add its counts into these totals.
113 * We won't ever get here for the group leader, since it
114 * will have been the last reference on the signal_struct.
116 sig->utime = cputime_add(sig->utime, tsk->utime);
117 sig->stime = cputime_add(sig->stime, tsk->stime);
118 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
119 sig->min_flt += tsk->min_flt;
120 sig->maj_flt += tsk->maj_flt;
121 sig->nvcsw += tsk->nvcsw;
122 sig->nivcsw += tsk->nivcsw;
123 sig->inblock += task_io_get_inblock(tsk);
124 sig->oublock += task_io_get_oublock(tsk);
125 task_io_accounting_add(&sig->ioac, &tsk->ioac);
126 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
130 __unhash_process(tsk, group_dead);
133 * Do this under ->siglock, we can race with another thread
134 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
136 flush_sigqueue(&tsk->pending);
138 spin_unlock(&sighand->siglock);
140 __cleanup_sighand(sighand);
141 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
143 flush_sigqueue(&sig->shared_pending);
148 static void delayed_put_task_struct(struct rcu_head *rhp)
150 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
152 perf_event_delayed_put(tsk);
153 trace_sched_process_free(tsk);
154 put_task_struct(tsk);
158 void release_task(struct task_struct * p)
160 struct task_struct *leader;
163 tracehook_prepare_release_task(p);
164 /* don't need to get the RCU readlock here - the process is dead and
165 * can't be modifying its own credentials. But shut RCU-lockdep up */
167 atomic_dec(&__task_cred(p)->user->processes);
172 write_lock_irq(&tasklist_lock);
173 tracehook_finish_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 * This maintains the invariant that release_task()
198 * only runs on a task in EXIT_DEAD, just for sanity.
201 leader->exit_state = EXIT_DEAD;
204 write_unlock_irq(&tasklist_lock);
206 call_rcu(&p->rcu, delayed_put_task_struct);
209 if (unlikely(zap_leader))
214 * This checks not only the pgrp, but falls back on the pid if no
215 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
218 * The caller must hold rcu lock or the tasklist lock.
220 struct pid *session_of_pgrp(struct pid *pgrp)
222 struct task_struct *p;
223 struct pid *sid = NULL;
225 p = pid_task(pgrp, PIDTYPE_PGID);
227 p = pid_task(pgrp, PIDTYPE_PID);
229 sid = task_session(p);
235 * Determine if a process group is "orphaned", according to the POSIX
236 * definition in 2.2.2.52. Orphaned process groups are not to be affected
237 * by terminal-generated stop signals. Newly orphaned process groups are
238 * to receive a SIGHUP and a SIGCONT.
240 * "I ask you, have you ever known what it is to be an orphan?"
242 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
244 struct task_struct *p;
246 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
247 if ((p == ignored_task) ||
248 (p->exit_state && thread_group_empty(p)) ||
249 is_global_init(p->real_parent))
252 if (task_pgrp(p->real_parent) != pgrp &&
253 task_session(p->real_parent) == task_session(p))
255 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
260 int is_current_pgrp_orphaned(void)
264 read_lock(&tasklist_lock);
265 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
266 read_unlock(&tasklist_lock);
271 static int has_stopped_jobs(struct pid *pgrp)
274 struct task_struct *p;
276 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
277 if (!task_is_stopped(p))
281 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
286 * Check to see if any process groups have become orphaned as
287 * a result of our exiting, and if they have any stopped jobs,
288 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
291 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
293 struct pid *pgrp = task_pgrp(tsk);
294 struct task_struct *ignored_task = tsk;
297 /* exit: our father is in a different pgrp than
298 * we are and we were the only connection outside.
300 parent = tsk->real_parent;
302 /* reparent: our child is in a different pgrp than
303 * we are, and it was the only connection outside.
307 if (task_pgrp(parent) != pgrp &&
308 task_session(parent) == task_session(tsk) &&
309 will_become_orphaned_pgrp(pgrp, ignored_task) &&
310 has_stopped_jobs(pgrp)) {
311 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
312 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
317 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
319 * If a kernel thread is launched as a result of a system call, or if
320 * it ever exits, it should generally reparent itself to kthreadd so it
321 * isn't in the way of other processes and is correctly cleaned up on exit.
323 * The various task state such as scheduling policy and priority may have
324 * been inherited from a user process, so we reset them to sane values here.
326 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
328 static void reparent_to_kthreadd(void)
330 write_lock_irq(&tasklist_lock);
332 ptrace_unlink(current);
333 /* Reparent to init */
334 current->real_parent = current->parent = kthreadd_task;
335 list_move_tail(¤t->sibling, ¤t->real_parent->children);
337 /* Set the exit signal to SIGCHLD so we signal init on exit */
338 current->exit_signal = SIGCHLD;
340 if (task_nice(current) < 0)
341 set_user_nice(current, 0);
345 memcpy(current->signal->rlim, init_task.signal->rlim,
346 sizeof(current->signal->rlim));
348 atomic_inc(&init_cred.usage);
349 commit_creds(&init_cred);
350 write_unlock_irq(&tasklist_lock);
353 void __set_special_pids(struct pid *pid)
355 struct task_struct *curr = current->group_leader;
357 if (task_session(curr) != pid)
358 change_pid(curr, PIDTYPE_SID, pid);
360 if (task_pgrp(curr) != pid)
361 change_pid(curr, PIDTYPE_PGID, pid);
364 static void set_special_pids(struct pid *pid)
366 write_lock_irq(&tasklist_lock);
367 __set_special_pids(pid);
368 write_unlock_irq(&tasklist_lock);
372 * Let kernel threads use this to say that they allow a certain signal.
373 * Must not be used if kthread was cloned with CLONE_SIGHAND.
375 int allow_signal(int sig)
377 if (!valid_signal(sig) || sig < 1)
380 spin_lock_irq(¤t->sighand->siglock);
381 /* This is only needed for daemonize()'ed kthreads */
382 sigdelset(¤t->blocked, sig);
384 * Kernel threads handle their own signals. Let the signal code
385 * know it'll be handled, so that they don't get converted to
386 * SIGKILL or just silently dropped.
388 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, ...)
420 va_start(args, name);
421 vsnprintf(current->comm, sizeof(current->comm), name, args);
425 * If we were started as result of loading a module, close all of the
426 * user space pages. We don't need them, and if we didn't close them
427 * they would be locked into memory.
431 * We don't want to have TIF_FREEZE set if the system-wide hibernation
432 * or suspend transition begins right now.
434 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
436 if (current->nsproxy != &init_nsproxy) {
437 get_nsproxy(&init_nsproxy);
438 switch_task_namespaces(current, &init_nsproxy);
440 set_special_pids(&init_struct_pid);
441 proc_clear_tty(current);
443 /* Block and flush all signals */
444 sigfillset(&blocked);
445 sigprocmask(SIG_BLOCK, &blocked, NULL);
446 flush_signals(current);
448 /* Become as one with the init task */
450 daemonize_fs_struct();
452 current->files = init_task.files;
453 atomic_inc(¤t->files->count);
455 reparent_to_kthreadd();
458 EXPORT_SYMBOL(daemonize);
460 static void close_files(struct files_struct * files)
468 * It is safe to dereference the fd table without RCU or
469 * ->file_lock because this is the last reference to the
470 * files structure. But use RCU to shut RCU-lockdep up.
473 fdt = files_fdtable(files);
478 if (i >= fdt->max_fds)
480 set = fdt->open_fds->fds_bits[j++];
483 struct file * file = xchg(&fdt->fd[i], NULL);
485 filp_close(file, files);
495 struct files_struct *get_files_struct(struct task_struct *task)
497 struct files_struct *files;
502 atomic_inc(&files->count);
508 void put_files_struct(struct files_struct *files)
512 if (atomic_dec_and_test(&files->count)) {
515 * Free the fd and fdset arrays if we expanded them.
516 * If the fdtable was embedded, pass files for freeing
517 * at the end of the RCU grace period. Otherwise,
518 * you can free files immediately.
521 fdt = files_fdtable(files);
522 if (fdt != &files->fdtab)
523 kmem_cache_free(files_cachep, files);
529 void reset_files_struct(struct files_struct *files)
531 struct task_struct *tsk = current;
532 struct files_struct *old;
538 put_files_struct(old);
541 void exit_files(struct task_struct *tsk)
543 struct files_struct * files = tsk->files;
549 put_files_struct(files);
553 #ifdef CONFIG_MM_OWNER
555 * Task p is exiting and it owned mm, lets find a new owner for it
558 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
561 * If there are other users of the mm and the owner (us) is exiting
562 * we need to find a new owner to take on the responsibility.
564 if (atomic_read(&mm->mm_users) <= 1)
571 void mm_update_next_owner(struct mm_struct *mm)
573 struct task_struct *c, *g, *p = current;
576 if (!mm_need_new_owner(mm, p))
579 read_lock(&tasklist_lock);
581 * Search in the children
583 list_for_each_entry(c, &p->children, sibling) {
585 goto assign_new_owner;
589 * Search in the siblings
591 list_for_each_entry(c, &p->real_parent->children, sibling) {
593 goto assign_new_owner;
597 * Search through everything else. We should not get
600 do_each_thread(g, c) {
602 goto assign_new_owner;
603 } while_each_thread(g, c);
605 read_unlock(&tasklist_lock);
607 * We found no owner yet mm_users > 1: this implies that we are
608 * most likely racing with swapoff (try_to_unuse()) or /proc or
609 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
618 * The task_lock protects c->mm from changing.
619 * We always want mm->owner->mm == mm
623 * Delay read_unlock() till we have the task_lock()
624 * to ensure that c does not slip away underneath us
626 read_unlock(&tasklist_lock);
636 #endif /* CONFIG_MM_OWNER */
639 * Turn us into a lazy TLB process if we
642 static void exit_mm(struct task_struct * tsk)
644 struct mm_struct *mm = tsk->mm;
645 struct core_state *core_state;
651 * Serialize with any possible pending coredump.
652 * We must hold mmap_sem around checking core_state
653 * and clearing tsk->mm. The core-inducing thread
654 * will increment ->nr_threads for each thread in the
655 * group with ->mm != NULL.
657 down_read(&mm->mmap_sem);
658 core_state = mm->core_state;
660 struct core_thread self;
661 up_read(&mm->mmap_sem);
664 self.next = xchg(&core_state->dumper.next, &self);
666 * Implies mb(), the result of xchg() must be visible
667 * to core_state->dumper.
669 if (atomic_dec_and_test(&core_state->nr_threads))
670 complete(&core_state->startup);
673 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
674 if (!self.task) /* see coredump_finish() */
678 __set_task_state(tsk, TASK_RUNNING);
679 down_read(&mm->mmap_sem);
681 atomic_inc(&mm->mm_count);
682 BUG_ON(mm != tsk->active_mm);
683 /* more a memory barrier than a real lock */
686 up_read(&mm->mmap_sem);
687 enter_lazy_tlb(mm, current);
688 /* We don't want this task to be frozen prematurely */
689 clear_freeze_flag(tsk);
691 mm_update_next_owner(mm);
696 * When we die, we re-parent all our children.
697 * Try to give them to another thread in our thread
698 * group, and if no such member exists, give it to
699 * the child reaper process (ie "init") in our pid
702 static struct task_struct *find_new_reaper(struct task_struct *father)
704 struct pid_namespace *pid_ns = task_active_pid_ns(father);
705 struct task_struct *thread;
708 while_each_thread(father, thread) {
709 if (thread->flags & PF_EXITING)
711 if (unlikely(pid_ns->child_reaper == father))
712 pid_ns->child_reaper = thread;
716 if (unlikely(pid_ns->child_reaper == father)) {
717 write_unlock_irq(&tasklist_lock);
718 if (unlikely(pid_ns == &init_pid_ns))
719 panic("Attempted to kill init!");
721 zap_pid_ns_processes(pid_ns);
722 write_lock_irq(&tasklist_lock);
724 * We can not clear ->child_reaper or leave it alone.
725 * There may by stealth EXIT_DEAD tasks on ->children,
726 * forget_original_parent() must move them somewhere.
728 pid_ns->child_reaper = init_pid_ns.child_reaper;
731 return pid_ns->child_reaper;
735 * Any that need to be release_task'd are put on the @dead list.
737 static void reparent_leader(struct task_struct *father, struct task_struct *p,
738 struct list_head *dead)
740 list_move_tail(&p->sibling, &p->real_parent->children);
742 if (task_detached(p))
745 * If this is a threaded reparent there is no need to
746 * notify anyone anything has happened.
748 if (same_thread_group(p->real_parent, father))
751 /* We don't want people slaying init. */
752 p->exit_signal = SIGCHLD;
754 /* If it has exited notify the new parent about this child's death. */
755 if (!task_ptrace(p) &&
756 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
757 do_notify_parent(p, p->exit_signal);
758 if (task_detached(p)) {
759 p->exit_state = EXIT_DEAD;
760 list_move_tail(&p->sibling, dead);
764 kill_orphaned_pgrp(p, father);
767 static void forget_original_parent(struct task_struct *father)
769 struct task_struct *p, *n, *reaper;
770 LIST_HEAD(dead_children);
772 write_lock_irq(&tasklist_lock);
774 * Note that exit_ptrace() and find_new_reaper() might
775 * drop tasklist_lock and reacquire it.
778 reaper = find_new_reaper(father);
780 list_for_each_entry_safe(p, n, &father->children, sibling) {
781 struct task_struct *t = p;
783 t->real_parent = reaper;
784 if (t->parent == father) {
785 BUG_ON(task_ptrace(t));
786 t->parent = t->real_parent;
788 if (t->pdeath_signal)
789 group_send_sig_info(t->pdeath_signal,
791 } while_each_thread(p, t);
792 reparent_leader(father, p, &dead_children);
794 write_unlock_irq(&tasklist_lock);
796 BUG_ON(!list_empty(&father->children));
798 list_for_each_entry_safe(p, n, &dead_children, sibling) {
799 list_del_init(&p->sibling);
805 * Send signals to all our closest relatives so that they know
806 * to properly mourn us..
808 static void exit_notify(struct task_struct *tsk, int group_dead)
814 * This does two things:
816 * A. Make init inherit all the child processes
817 * B. Check to see if any process groups have become orphaned
818 * as a result of our exiting, and if they have any stopped
819 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
821 forget_original_parent(tsk);
822 exit_task_namespaces(tsk);
824 write_lock_irq(&tasklist_lock);
826 kill_orphaned_pgrp(tsk->group_leader, NULL);
828 /* Let father know we died
830 * Thread signals are configurable, but you aren't going to use
831 * that to send signals to arbitary processes.
832 * That stops right now.
834 * If the parent exec id doesn't match the exec id we saved
835 * when we started then we know the parent has changed security
838 * If our self_exec id doesn't match our parent_exec_id then
839 * we have changed execution domain as these two values started
840 * the same after a fork.
842 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
843 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
844 tsk->self_exec_id != tsk->parent_exec_id))
845 tsk->exit_signal = SIGCHLD;
847 signal = tracehook_notify_death(tsk, &cookie, group_dead);
849 signal = do_notify_parent(tsk, signal);
851 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
853 /* mt-exec, de_thread() is waiting for group leader */
854 if (unlikely(tsk->signal->notify_count < 0))
855 wake_up_process(tsk->signal->group_exit_task);
856 write_unlock_irq(&tasklist_lock);
858 tracehook_report_death(tsk, signal, cookie, group_dead);
860 /* If the process is dead, release it - nobody will wait for it */
861 if (signal == DEATH_REAP)
865 #ifdef CONFIG_DEBUG_STACK_USAGE
866 static void check_stack_usage(void)
868 static DEFINE_SPINLOCK(low_water_lock);
869 static int lowest_to_date = THREAD_SIZE;
872 free = stack_not_used(current);
874 if (free >= lowest_to_date)
877 spin_lock(&low_water_lock);
878 if (free < lowest_to_date) {
879 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
881 current->comm, free);
882 lowest_to_date = free;
884 spin_unlock(&low_water_lock);
887 static inline void check_stack_usage(void) {}
890 NORET_TYPE void do_exit(long code)
892 struct task_struct *tsk = current;
895 profile_task_exit(tsk);
897 WARN_ON(atomic_read(&tsk->fs_excl));
899 if (unlikely(in_interrupt()))
900 panic("Aiee, killing interrupt handler!");
901 if (unlikely(!tsk->pid))
902 panic("Attempted to kill the idle task!");
904 tracehook_report_exit(&code);
906 validate_creds_for_do_exit(tsk);
909 * We're taking recursive faults here in do_exit. Safest is to just
910 * leave this task alone and wait for reboot.
912 if (unlikely(tsk->flags & PF_EXITING)) {
914 "Fixing recursive fault but reboot is needed!\n");
916 * We can do this unlocked here. The futex code uses
917 * this flag just to verify whether the pi state
918 * cleanup has been done or not. In the worst case it
919 * loops once more. We pretend that the cleanup was
920 * done as there is no way to return. Either the
921 * OWNER_DIED bit is set by now or we push the blocked
922 * task into the wait for ever nirwana as well.
924 tsk->flags |= PF_EXITPIDONE;
925 set_current_state(TASK_UNINTERRUPTIBLE);
931 exit_signals(tsk); /* sets PF_EXITING */
933 * tsk->flags are checked in the futex code to protect against
934 * an exiting task cleaning up the robust pi futexes.
937 raw_spin_unlock_wait(&tsk->pi_lock);
939 if (unlikely(in_atomic()))
940 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
941 current->comm, task_pid_nr(current),
944 acct_update_integrals(tsk);
945 /* sync mm's RSS info before statistics gathering */
947 sync_mm_rss(tsk, tsk->mm);
948 group_dead = atomic_dec_and_test(&tsk->signal->live);
950 hrtimer_cancel(&tsk->signal->real_timer);
951 exit_itimers(tsk->signal);
953 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
955 acct_collect(code, group_dead);
958 if (unlikely(tsk->audit_context))
961 tsk->exit_code = code;
962 taskstats_exit(tsk, group_dead);
968 trace_sched_process_exit(tsk);
978 disassociate_ctty(1);
980 module_put(task_thread_info(tsk)->exec_domain->module);
982 proc_exit_connector(tsk);
985 * FIXME: do that only when needed, using sched_exit tracepoint
987 flush_ptrace_hw_breakpoint(tsk);
989 * Flush inherited counters to the parent - before the parent
990 * gets woken up by child-exit notifications.
992 perf_event_exit_task(tsk);
994 exit_notify(tsk, group_dead);
997 mpol_put(tsk->mempolicy);
998 tsk->mempolicy = NULL;
1002 if (unlikely(current->pi_state_cache))
1003 kfree(current->pi_state_cache);
1006 * Make sure we are holding no locks:
1008 debug_check_no_locks_held(tsk);
1010 * We can do this unlocked here. The futex code uses this flag
1011 * just to verify whether the pi state cleanup has been done
1012 * or not. In the worst case it loops once more.
1014 tsk->flags |= PF_EXITPIDONE;
1016 if (tsk->io_context)
1017 exit_io_context(tsk);
1019 if (tsk->splice_pipe)
1020 __free_pipe_info(tsk->splice_pipe);
1022 validate_creds_for_do_exit(tsk);
1026 /* causes final put_task_struct in finish_task_switch(). */
1027 tsk->state = TASK_DEAD;
1030 /* Avoid "noreturn function does return". */
1032 cpu_relax(); /* For when BUG is null */
1035 EXPORT_SYMBOL_GPL(do_exit);
1037 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1045 EXPORT_SYMBOL(complete_and_exit);
1047 SYSCALL_DEFINE1(exit, int, error_code)
1049 do_exit((error_code&0xff)<<8);
1053 * Take down every thread in the group. This is called by fatal signals
1054 * as well as by sys_exit_group (below).
1057 do_group_exit(int exit_code)
1059 struct signal_struct *sig = current->signal;
1061 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1063 if (signal_group_exit(sig))
1064 exit_code = sig->group_exit_code;
1065 else if (!thread_group_empty(current)) {
1066 struct sighand_struct *const sighand = current->sighand;
1067 spin_lock_irq(&sighand->siglock);
1068 if (signal_group_exit(sig))
1069 /* Another thread got here before we took the lock. */
1070 exit_code = sig->group_exit_code;
1072 sig->group_exit_code = exit_code;
1073 sig->flags = SIGNAL_GROUP_EXIT;
1074 zap_other_threads(current);
1076 spin_unlock_irq(&sighand->siglock);
1084 * this kills every thread in the thread group. Note that any externally
1085 * wait4()-ing process will get the correct exit code - even if this
1086 * thread is not the thread group leader.
1088 SYSCALL_DEFINE1(exit_group, int, error_code)
1090 do_group_exit((error_code & 0xff) << 8);
1096 enum pid_type wo_type;
1100 struct siginfo __user *wo_info;
1101 int __user *wo_stat;
1102 struct rusage __user *wo_rusage;
1104 wait_queue_t child_wait;
1109 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1111 if (type != PIDTYPE_PID)
1112 task = task->group_leader;
1113 return task->pids[type].pid;
1116 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1118 return wo->wo_type == PIDTYPE_MAX ||
1119 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1122 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1124 if (!eligible_pid(wo, p))
1126 /* Wait for all children (clone and not) if __WALL is set;
1127 * otherwise, wait for clone children *only* if __WCLONE is
1128 * set; otherwise, wait for non-clone children *only*. (Note:
1129 * A "clone" child here is one that reports to its parent
1130 * using a signal other than SIGCHLD.) */
1131 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1132 && !(wo->wo_flags & __WALL))
1138 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1139 pid_t pid, uid_t uid, int why, int status)
1141 struct siginfo __user *infop;
1142 int retval = wo->wo_rusage
1143 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1146 infop = wo->wo_info;
1149 retval = put_user(SIGCHLD, &infop->si_signo);
1151 retval = put_user(0, &infop->si_errno);
1153 retval = put_user((short)why, &infop->si_code);
1155 retval = put_user(pid, &infop->si_pid);
1157 retval = put_user(uid, &infop->si_uid);
1159 retval = put_user(status, &infop->si_status);
1167 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1168 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1169 * the lock and this task is uninteresting. If we return nonzero, we have
1170 * released the lock and the system call should return.
1172 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1174 unsigned long state;
1175 int retval, status, traced;
1176 pid_t pid = task_pid_vnr(p);
1177 uid_t uid = __task_cred(p)->uid;
1178 struct siginfo __user *infop;
1180 if (!likely(wo->wo_flags & WEXITED))
1183 if (unlikely(wo->wo_flags & WNOWAIT)) {
1184 int exit_code = p->exit_code;
1188 read_unlock(&tasklist_lock);
1189 if ((exit_code & 0x7f) == 0) {
1191 status = exit_code >> 8;
1193 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1194 status = exit_code & 0x7f;
1196 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1200 * Try to move the task's state to DEAD
1201 * only one thread is allowed to do this:
1203 state = xchg(&p->exit_state, EXIT_DEAD);
1204 if (state != EXIT_ZOMBIE) {
1205 BUG_ON(state != EXIT_DEAD);
1209 traced = ptrace_reparented(p);
1211 * It can be ptraced but not reparented, check
1212 * !task_detached() to filter out sub-threads.
1214 if (likely(!traced) && likely(!task_detached(p))) {
1215 struct signal_struct *psig;
1216 struct signal_struct *sig;
1217 unsigned long maxrss;
1218 cputime_t tgutime, tgstime;
1221 * The resource counters for the group leader are in its
1222 * own task_struct. Those for dead threads in the group
1223 * are in its signal_struct, as are those for the child
1224 * processes it has previously reaped. All these
1225 * accumulate in the parent's signal_struct c* fields.
1227 * We don't bother to take a lock here to protect these
1228 * p->signal fields, because they are only touched by
1229 * __exit_signal, which runs with tasklist_lock
1230 * write-locked anyway, and so is excluded here. We do
1231 * need to protect the access to parent->signal fields,
1232 * as other threads in the parent group can be right
1233 * here reaping other children at the same time.
1235 * We use thread_group_times() to get times for the thread
1236 * group, which consolidates times for all threads in the
1237 * group including the group leader.
1239 thread_group_times(p, &tgutime, &tgstime);
1240 spin_lock_irq(&p->real_parent->sighand->siglock);
1241 psig = p->real_parent->signal;
1244 cputime_add(psig->cutime,
1245 cputime_add(tgutime,
1248 cputime_add(psig->cstime,
1249 cputime_add(tgstime,
1252 cputime_add(psig->cgtime,
1253 cputime_add(p->gtime,
1254 cputime_add(sig->gtime,
1257 p->min_flt + sig->min_flt + sig->cmin_flt;
1259 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1261 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1263 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1265 task_io_get_inblock(p) +
1266 sig->inblock + sig->cinblock;
1268 task_io_get_oublock(p) +
1269 sig->oublock + sig->coublock;
1270 maxrss = max(sig->maxrss, sig->cmaxrss);
1271 if (psig->cmaxrss < maxrss)
1272 psig->cmaxrss = maxrss;
1273 task_io_accounting_add(&psig->ioac, &p->ioac);
1274 task_io_accounting_add(&psig->ioac, &sig->ioac);
1275 spin_unlock_irq(&p->real_parent->sighand->siglock);
1279 * Now we are sure this task is interesting, and no other
1280 * thread can reap it because we set its state to EXIT_DEAD.
1282 read_unlock(&tasklist_lock);
1284 retval = wo->wo_rusage
1285 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1286 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1287 ? p->signal->group_exit_code : p->exit_code;
1288 if (!retval && wo->wo_stat)
1289 retval = put_user(status, wo->wo_stat);
1291 infop = wo->wo_info;
1292 if (!retval && infop)
1293 retval = put_user(SIGCHLD, &infop->si_signo);
1294 if (!retval && infop)
1295 retval = put_user(0, &infop->si_errno);
1296 if (!retval && infop) {
1299 if ((status & 0x7f) == 0) {
1303 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1306 retval = put_user((short)why, &infop->si_code);
1308 retval = put_user(status, &infop->si_status);
1310 if (!retval && infop)
1311 retval = put_user(pid, &infop->si_pid);
1312 if (!retval && infop)
1313 retval = put_user(uid, &infop->si_uid);
1318 write_lock_irq(&tasklist_lock);
1319 /* We dropped tasklist, ptracer could die and untrace */
1322 * If this is not a detached task, notify the parent.
1323 * If it's still not detached after that, don't release
1326 if (!task_detached(p)) {
1327 do_notify_parent(p, p->exit_signal);
1328 if (!task_detached(p)) {
1329 p->exit_state = EXIT_ZOMBIE;
1333 write_unlock_irq(&tasklist_lock);
1341 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1344 if (task_is_stopped_or_traced(p))
1345 return &p->exit_code;
1347 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1348 return &p->signal->group_exit_code;
1354 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1355 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1356 * the lock and this task is uninteresting. If we return nonzero, we have
1357 * released the lock and the system call should return.
1359 static int wait_task_stopped(struct wait_opts *wo,
1360 int ptrace, struct task_struct *p)
1362 struct siginfo __user *infop;
1363 int retval, exit_code, *p_code, why;
1364 uid_t uid = 0; /* unneeded, required by compiler */
1368 * Traditionally we see ptrace'd stopped tasks regardless of options.
1370 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1374 spin_lock_irq(&p->sighand->siglock);
1376 p_code = task_stopped_code(p, ptrace);
1377 if (unlikely(!p_code))
1380 exit_code = *p_code;
1384 if (!unlikely(wo->wo_flags & WNOWAIT))
1389 spin_unlock_irq(&p->sighand->siglock);
1394 * Now we are pretty sure this task is interesting.
1395 * Make sure it doesn't get reaped out from under us while we
1396 * give up the lock and then examine it below. We don't want to
1397 * keep holding onto the tasklist_lock while we call getrusage and
1398 * possibly take page faults for user memory.
1401 pid = task_pid_vnr(p);
1402 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1403 read_unlock(&tasklist_lock);
1405 if (unlikely(wo->wo_flags & WNOWAIT))
1406 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1408 retval = wo->wo_rusage
1409 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1410 if (!retval && wo->wo_stat)
1411 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1413 infop = wo->wo_info;
1414 if (!retval && infop)
1415 retval = put_user(SIGCHLD, &infop->si_signo);
1416 if (!retval && infop)
1417 retval = put_user(0, &infop->si_errno);
1418 if (!retval && infop)
1419 retval = put_user((short)why, &infop->si_code);
1420 if (!retval && infop)
1421 retval = put_user(exit_code, &infop->si_status);
1422 if (!retval && infop)
1423 retval = put_user(pid, &infop->si_pid);
1424 if (!retval && infop)
1425 retval = put_user(uid, &infop->si_uid);
1435 * Handle do_wait work for one task in a live, non-stopped state.
1436 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1437 * the lock and this task is uninteresting. If we return nonzero, we have
1438 * released the lock and the system call should return.
1440 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1446 if (!unlikely(wo->wo_flags & WCONTINUED))
1449 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1452 spin_lock_irq(&p->sighand->siglock);
1453 /* Re-check with the lock held. */
1454 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1455 spin_unlock_irq(&p->sighand->siglock);
1458 if (!unlikely(wo->wo_flags & WNOWAIT))
1459 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1461 spin_unlock_irq(&p->sighand->siglock);
1463 pid = task_pid_vnr(p);
1465 read_unlock(&tasklist_lock);
1468 retval = wo->wo_rusage
1469 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1471 if (!retval && wo->wo_stat)
1472 retval = put_user(0xffff, wo->wo_stat);
1476 retval = wait_noreap_copyout(wo, p, pid, uid,
1477 CLD_CONTINUED, SIGCONT);
1478 BUG_ON(retval == 0);
1485 * Consider @p for a wait by @parent.
1487 * -ECHILD should be in ->notask_error before the first call.
1488 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1489 * Returns zero if the search for a child should continue;
1490 * then ->notask_error is 0 if @p is an eligible child,
1491 * or another error from security_task_wait(), or still -ECHILD.
1493 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1494 struct task_struct *p)
1496 int ret = eligible_child(wo, p);
1500 ret = security_task_wait(p);
1501 if (unlikely(ret < 0)) {
1503 * If we have not yet seen any eligible child,
1504 * then let this error code replace -ECHILD.
1505 * A permission error will give the user a clue
1506 * to look for security policy problems, rather
1507 * than for mysterious wait bugs.
1509 if (wo->notask_error)
1510 wo->notask_error = ret;
1514 if (likely(!ptrace) && unlikely(task_ptrace(p))) {
1516 * This child is hidden by ptrace.
1517 * We aren't allowed to see it now, but eventually we will.
1519 wo->notask_error = 0;
1523 if (p->exit_state == EXIT_DEAD)
1527 * We don't reap group leaders with subthreads.
1529 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1530 return wait_task_zombie(wo, p);
1533 * It's stopped or running now, so it might
1534 * later continue, exit, or stop again.
1536 wo->notask_error = 0;
1538 if (task_stopped_code(p, ptrace))
1539 return wait_task_stopped(wo, ptrace, p);
1541 return wait_task_continued(wo, p);
1545 * Do the work of do_wait() for one thread in the group, @tsk.
1547 * -ECHILD should be in ->notask_error before the first call.
1548 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1549 * Returns zero if the search for a child should continue; then
1550 * ->notask_error is 0 if there were any eligible children,
1551 * or another error from security_task_wait(), or still -ECHILD.
1553 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1555 struct task_struct *p;
1557 list_for_each_entry(p, &tsk->children, sibling) {
1558 int ret = wait_consider_task(wo, 0, p);
1566 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1568 struct task_struct *p;
1570 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1571 int ret = wait_consider_task(wo, 1, p);
1579 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1580 int sync, void *key)
1582 struct wait_opts *wo = container_of(wait, struct wait_opts,
1584 struct task_struct *p = key;
1586 if (!eligible_pid(wo, p))
1589 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1592 return default_wake_function(wait, mode, sync, key);
1595 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1597 __wake_up_sync_key(&parent->signal->wait_chldexit,
1598 TASK_INTERRUPTIBLE, 1, p);
1601 static long do_wait(struct wait_opts *wo)
1603 struct task_struct *tsk;
1606 trace_sched_process_wait(wo->wo_pid);
1608 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1609 wo->child_wait.private = current;
1610 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1613 * If there is nothing that can match our critiera just get out.
1614 * We will clear ->notask_error to zero if we see any child that
1615 * might later match our criteria, even if we are not able to reap
1618 wo->notask_error = -ECHILD;
1619 if ((wo->wo_type < PIDTYPE_MAX) &&
1620 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1623 set_current_state(TASK_INTERRUPTIBLE);
1624 read_lock(&tasklist_lock);
1627 retval = do_wait_thread(wo, tsk);
1631 retval = ptrace_do_wait(wo, tsk);
1635 if (wo->wo_flags & __WNOTHREAD)
1637 } while_each_thread(current, tsk);
1638 read_unlock(&tasklist_lock);
1641 retval = wo->notask_error;
1642 if (!retval && !(wo->wo_flags & WNOHANG)) {
1643 retval = -ERESTARTSYS;
1644 if (!signal_pending(current)) {
1650 __set_current_state(TASK_RUNNING);
1651 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1655 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1656 infop, int, options, struct rusage __user *, ru)
1658 struct wait_opts wo;
1659 struct pid *pid = NULL;
1663 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1665 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1678 type = PIDTYPE_PGID;
1686 if (type < PIDTYPE_MAX)
1687 pid = find_get_pid(upid);
1691 wo.wo_flags = options;
1701 * For a WNOHANG return, clear out all the fields
1702 * we would set so the user can easily tell the
1706 ret = put_user(0, &infop->si_signo);
1708 ret = put_user(0, &infop->si_errno);
1710 ret = put_user(0, &infop->si_code);
1712 ret = put_user(0, &infop->si_pid);
1714 ret = put_user(0, &infop->si_uid);
1716 ret = put_user(0, &infop->si_status);
1721 /* avoid REGPARM breakage on x86: */
1722 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1726 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1727 int, options, struct rusage __user *, ru)
1729 struct wait_opts wo;
1730 struct pid *pid = NULL;
1734 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1735 __WNOTHREAD|__WCLONE|__WALL))
1740 else if (upid < 0) {
1741 type = PIDTYPE_PGID;
1742 pid = find_get_pid(-upid);
1743 } else if (upid == 0) {
1744 type = PIDTYPE_PGID;
1745 pid = get_task_pid(current, PIDTYPE_PGID);
1746 } else /* upid > 0 */ {
1748 pid = find_get_pid(upid);
1753 wo.wo_flags = options | WEXITED;
1755 wo.wo_stat = stat_addr;
1760 /* avoid REGPARM breakage on x86: */
1761 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1765 #ifdef __ARCH_WANT_SYS_WAITPID
1768 * sys_waitpid() remains for compatibility. waitpid() should be
1769 * implemented by calling sys_wait4() from libc.a.
1771 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1773 return sys_wait4(pid, stat_addr, options, NULL);