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)
64 detach_pid(p, PIDTYPE_PID);
65 if (thread_group_leader(p)) {
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 struct sighand_struct *sighand;
85 BUG_ON(!atomic_read(&sig->count));
87 sighand = rcu_dereference_check(tsk->sighand,
88 rcu_read_lock_held() ||
89 lockdep_tasklist_lock_is_held());
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 task_io_accounting_add(&sig->ioac, &tsk->ioac);
125 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
126 sig = NULL; /* Marker for below. */
129 __unhash_process(tsk);
132 * Do this under ->siglock, we can race with another thread
133 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
135 flush_sigqueue(&tsk->pending);
139 spin_unlock(&sighand->siglock);
141 __cleanup_sighand(sighand);
142 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
144 flush_sigqueue(&sig->shared_pending);
145 taskstats_tgid_free(sig);
147 * Make sure ->signal can't go away under rq->lock,
148 * see account_group_exec_runtime().
150 task_rq_unlock_wait(tsk);
151 __cleanup_signal(sig);
155 static void delayed_put_task_struct(struct rcu_head *rhp)
157 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
159 #ifdef CONFIG_PERF_EVENTS
160 WARN_ON_ONCE(tsk->perf_event_ctxp);
162 trace_sched_process_free(tsk);
163 put_task_struct(tsk);
167 void release_task(struct task_struct * p)
169 struct task_struct *leader;
172 tracehook_prepare_release_task(p);
173 /* don't need to get the RCU readlock here - the process is dead and
174 * can't be modifying its own credentials. But shut RCU-lockdep up */
176 atomic_dec(&__task_cred(p)->user->processes);
181 write_lock_irq(&tasklist_lock);
182 tracehook_finish_release_task(p);
186 * If we are the last non-leader member of the thread
187 * group, and the leader is zombie, then notify the
188 * group leader's parent process. (if it wants notification.)
191 leader = p->group_leader;
192 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
193 BUG_ON(task_detached(leader));
194 do_notify_parent(leader, leader->exit_signal);
196 * If we were the last child thread and the leader has
197 * exited already, and the leader's parent ignores SIGCHLD,
198 * then we are the one who should release the leader.
200 * do_notify_parent() will have marked it self-reaping in
203 zap_leader = task_detached(leader);
206 * This maintains the invariant that release_task()
207 * only runs on a task in EXIT_DEAD, just for sanity.
210 leader->exit_state = EXIT_DEAD;
213 write_unlock_irq(&tasklist_lock);
215 call_rcu(&p->rcu, delayed_put_task_struct);
218 if (unlikely(zap_leader))
223 * This checks not only the pgrp, but falls back on the pid if no
224 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
227 * The caller must hold rcu lock or the tasklist lock.
229 struct pid *session_of_pgrp(struct pid *pgrp)
231 struct task_struct *p;
232 struct pid *sid = NULL;
234 p = pid_task(pgrp, PIDTYPE_PGID);
236 p = pid_task(pgrp, PIDTYPE_PID);
238 sid = task_session(p);
244 * Determine if a process group is "orphaned", according to the POSIX
245 * definition in 2.2.2.52. Orphaned process groups are not to be affected
246 * by terminal-generated stop signals. Newly orphaned process groups are
247 * to receive a SIGHUP and a SIGCONT.
249 * "I ask you, have you ever known what it is to be an orphan?"
251 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
253 struct task_struct *p;
255 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
256 if ((p == ignored_task) ||
257 (p->exit_state && thread_group_empty(p)) ||
258 is_global_init(p->real_parent))
261 if (task_pgrp(p->real_parent) != pgrp &&
262 task_session(p->real_parent) == task_session(p))
264 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
269 int is_current_pgrp_orphaned(void)
273 read_lock(&tasklist_lock);
274 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
275 read_unlock(&tasklist_lock);
280 static int has_stopped_jobs(struct pid *pgrp)
283 struct task_struct *p;
285 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
286 if (!task_is_stopped(p))
290 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
295 * Check to see if any process groups have become orphaned as
296 * a result of our exiting, and if they have any stopped jobs,
297 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
300 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
302 struct pid *pgrp = task_pgrp(tsk);
303 struct task_struct *ignored_task = tsk;
306 /* exit: our father is in a different pgrp than
307 * we are and we were the only connection outside.
309 parent = tsk->real_parent;
311 /* reparent: our child is in a different pgrp than
312 * we are, and it was the only connection outside.
316 if (task_pgrp(parent) != pgrp &&
317 task_session(parent) == task_session(tsk) &&
318 will_become_orphaned_pgrp(pgrp, ignored_task) &&
319 has_stopped_jobs(pgrp)) {
320 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
321 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
326 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
328 * If a kernel thread is launched as a result of a system call, or if
329 * it ever exits, it should generally reparent itself to kthreadd so it
330 * isn't in the way of other processes and is correctly cleaned up on exit.
332 * The various task state such as scheduling policy and priority may have
333 * been inherited from a user process, so we reset them to sane values here.
335 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
337 static void reparent_to_kthreadd(void)
339 write_lock_irq(&tasklist_lock);
341 ptrace_unlink(current);
342 /* Reparent to init */
343 current->real_parent = current->parent = kthreadd_task;
344 list_move_tail(¤t->sibling, ¤t->real_parent->children);
346 /* Set the exit signal to SIGCHLD so we signal init on exit */
347 current->exit_signal = SIGCHLD;
349 if (task_nice(current) < 0)
350 set_user_nice(current, 0);
354 memcpy(current->signal->rlim, init_task.signal->rlim,
355 sizeof(current->signal->rlim));
357 atomic_inc(&init_cred.usage);
358 commit_creds(&init_cred);
359 write_unlock_irq(&tasklist_lock);
362 void __set_special_pids(struct pid *pid)
364 struct task_struct *curr = current->group_leader;
366 if (task_session(curr) != pid)
367 change_pid(curr, PIDTYPE_SID, pid);
369 if (task_pgrp(curr) != pid)
370 change_pid(curr, PIDTYPE_PGID, pid);
373 static void set_special_pids(struct pid *pid)
375 write_lock_irq(&tasklist_lock);
376 __set_special_pids(pid);
377 write_unlock_irq(&tasklist_lock);
381 * Let kernel threads use this to say that they allow a certain signal.
382 * Must not be used if kthread was cloned with CLONE_SIGHAND.
384 int allow_signal(int sig)
386 if (!valid_signal(sig) || sig < 1)
389 spin_lock_irq(¤t->sighand->siglock);
390 /* This is only needed for daemonize()'ed kthreads */
391 sigdelset(¤t->blocked, sig);
393 * Kernel threads handle their own signals. Let the signal code
394 * know it'll be handled, so that they don't get converted to
395 * SIGKILL or just silently dropped.
397 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
399 spin_unlock_irq(¤t->sighand->siglock);
403 EXPORT_SYMBOL(allow_signal);
405 int disallow_signal(int sig)
407 if (!valid_signal(sig) || sig < 1)
410 spin_lock_irq(¤t->sighand->siglock);
411 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
413 spin_unlock_irq(¤t->sighand->siglock);
417 EXPORT_SYMBOL(disallow_signal);
420 * Put all the gunge required to become a kernel thread without
421 * attached user resources in one place where it belongs.
424 void daemonize(const char *name, ...)
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 daemonize_fs_struct();
461 current->files = init_task.files;
462 atomic_inc(¤t->files->count);
464 reparent_to_kthreadd();
467 EXPORT_SYMBOL(daemonize);
469 static void close_files(struct files_struct * files)
477 * It is safe to dereference the fd table without RCU or
478 * ->file_lock because this is the last reference to the
479 * files structure. But use RCU to shut RCU-lockdep up.
482 fdt = files_fdtable(files);
487 if (i >= fdt->max_fds)
489 set = fdt->open_fds->fds_bits[j++];
492 struct file * file = xchg(&fdt->fd[i], NULL);
494 filp_close(file, files);
504 struct files_struct *get_files_struct(struct task_struct *task)
506 struct files_struct *files;
511 atomic_inc(&files->count);
517 void put_files_struct(struct files_struct *files)
521 if (atomic_dec_and_test(&files->count)) {
524 * Free the fd and fdset arrays if we expanded them.
525 * If the fdtable was embedded, pass files for freeing
526 * at the end of the RCU grace period. Otherwise,
527 * you can free files immediately.
530 fdt = files_fdtable(files);
531 if (fdt != &files->fdtab)
532 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 #ifdef CONFIG_MM_OWNER
564 * Task p is exiting and it owned mm, lets find a new owner for it
567 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
570 * If there are other users of the mm and the owner (us) is exiting
571 * we need to find a new owner to take on the responsibility.
573 if (atomic_read(&mm->mm_users) <= 1)
580 void mm_update_next_owner(struct mm_struct *mm)
582 struct task_struct *c, *g, *p = current;
585 if (!mm_need_new_owner(mm, p))
588 read_lock(&tasklist_lock);
590 * Search in the children
592 list_for_each_entry(c, &p->children, sibling) {
594 goto assign_new_owner;
598 * Search in the siblings
600 list_for_each_entry(c, &p->real_parent->children, sibling) {
602 goto assign_new_owner;
606 * Search through everything else. We should not get
609 do_each_thread(g, c) {
611 goto assign_new_owner;
612 } while_each_thread(g, c);
614 read_unlock(&tasklist_lock);
616 * We found no owner yet mm_users > 1: this implies that we are
617 * most likely racing with swapoff (try_to_unuse()) or /proc or
618 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
627 * The task_lock protects c->mm from changing.
628 * We always want mm->owner->mm == mm
632 * Delay read_unlock() till we have the task_lock()
633 * to ensure that c does not slip away underneath us
635 read_unlock(&tasklist_lock);
645 #endif /* CONFIG_MM_OWNER */
648 * Turn us into a lazy TLB process if we
651 static void exit_mm(struct task_struct * tsk)
653 struct mm_struct *mm = tsk->mm;
654 struct core_state *core_state;
660 * Serialize with any possible pending coredump.
661 * We must hold mmap_sem around checking core_state
662 * and clearing tsk->mm. The core-inducing thread
663 * will increment ->nr_threads for each thread in the
664 * group with ->mm != NULL.
666 down_read(&mm->mmap_sem);
667 core_state = mm->core_state;
669 struct core_thread self;
670 up_read(&mm->mmap_sem);
673 self.next = xchg(&core_state->dumper.next, &self);
675 * Implies mb(), the result of xchg() must be visible
676 * to core_state->dumper.
678 if (atomic_dec_and_test(&core_state->nr_threads))
679 complete(&core_state->startup);
682 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
683 if (!self.task) /* see coredump_finish() */
687 __set_task_state(tsk, TASK_RUNNING);
688 down_read(&mm->mmap_sem);
690 atomic_inc(&mm->mm_count);
691 BUG_ON(mm != tsk->active_mm);
692 /* more a memory barrier than a real lock */
695 up_read(&mm->mmap_sem);
696 enter_lazy_tlb(mm, current);
697 /* We don't want this task to be frozen prematurely */
698 clear_freeze_flag(tsk);
700 mm_update_next_owner(mm);
705 * When we die, we re-parent all our children.
706 * Try to give them to another thread in our thread
707 * group, and if no such member exists, give it to
708 * the child reaper process (ie "init") in our pid
711 static struct task_struct *find_new_reaper(struct task_struct *father)
713 struct pid_namespace *pid_ns = task_active_pid_ns(father);
714 struct task_struct *thread;
717 while_each_thread(father, thread) {
718 if (thread->flags & PF_EXITING)
720 if (unlikely(pid_ns->child_reaper == father))
721 pid_ns->child_reaper = thread;
725 if (unlikely(pid_ns->child_reaper == father)) {
726 write_unlock_irq(&tasklist_lock);
727 if (unlikely(pid_ns == &init_pid_ns))
728 panic("Attempted to kill init!");
730 zap_pid_ns_processes(pid_ns);
731 write_lock_irq(&tasklist_lock);
733 * We can not clear ->child_reaper or leave it alone.
734 * There may by stealth EXIT_DEAD tasks on ->children,
735 * forget_original_parent() must move them somewhere.
737 pid_ns->child_reaper = init_pid_ns.child_reaper;
740 return pid_ns->child_reaper;
744 * Any that need to be release_task'd are put on the @dead list.
746 static void reparent_leader(struct task_struct *father, struct task_struct *p,
747 struct list_head *dead)
749 list_move_tail(&p->sibling, &p->real_parent->children);
751 if (task_detached(p))
754 * If this is a threaded reparent there is no need to
755 * notify anyone anything has happened.
757 if (same_thread_group(p->real_parent, father))
760 /* We don't want people slaying init. */
761 p->exit_signal = SIGCHLD;
763 /* If it has exited notify the new parent about this child's death. */
764 if (!task_ptrace(p) &&
765 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
766 do_notify_parent(p, p->exit_signal);
767 if (task_detached(p)) {
768 p->exit_state = EXIT_DEAD;
769 list_move_tail(&p->sibling, dead);
773 kill_orphaned_pgrp(p, father);
776 static void forget_original_parent(struct task_struct *father)
778 struct task_struct *p, *n, *reaper;
779 LIST_HEAD(dead_children);
783 write_lock_irq(&tasklist_lock);
784 reaper = find_new_reaper(father);
786 list_for_each_entry_safe(p, n, &father->children, sibling) {
787 struct task_struct *t = p;
789 t->real_parent = reaper;
790 if (t->parent == father) {
791 BUG_ON(task_ptrace(t));
792 t->parent = t->real_parent;
794 if (t->pdeath_signal)
795 group_send_sig_info(t->pdeath_signal,
797 } while_each_thread(p, t);
798 reparent_leader(father, p, &dead_children);
800 write_unlock_irq(&tasklist_lock);
802 BUG_ON(!list_empty(&father->children));
804 list_for_each_entry_safe(p, n, &dead_children, sibling) {
805 list_del_init(&p->sibling);
811 * Send signals to all our closest relatives so that they know
812 * to properly mourn us..
814 static void exit_notify(struct task_struct *tsk, int group_dead)
820 * This does two things:
822 * A. Make init inherit all the child processes
823 * B. Check to see if any process groups have become orphaned
824 * as a result of our exiting, and if they have any stopped
825 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
827 forget_original_parent(tsk);
828 exit_task_namespaces(tsk);
830 write_lock_irq(&tasklist_lock);
832 kill_orphaned_pgrp(tsk->group_leader, NULL);
834 /* Let father know we died
836 * Thread signals are configurable, but you aren't going to use
837 * that to send signals to arbitary processes.
838 * That stops right now.
840 * If the parent exec id doesn't match the exec id we saved
841 * when we started then we know the parent has changed security
844 * If our self_exec id doesn't match our parent_exec_id then
845 * we have changed execution domain as these two values started
846 * the same after a fork.
848 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
849 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
850 tsk->self_exec_id != tsk->parent_exec_id))
851 tsk->exit_signal = SIGCHLD;
853 signal = tracehook_notify_death(tsk, &cookie, group_dead);
855 signal = do_notify_parent(tsk, signal);
857 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
859 /* mt-exec, de_thread() is waiting for us */
860 if (thread_group_leader(tsk) &&
861 tsk->signal->group_exit_task &&
862 tsk->signal->notify_count < 0)
863 wake_up_process(tsk->signal->group_exit_task);
865 write_unlock_irq(&tasklist_lock);
867 tracehook_report_death(tsk, signal, cookie, group_dead);
869 /* If the process is dead, release it - nobody will wait for it */
870 if (signal == DEATH_REAP)
874 #ifdef CONFIG_DEBUG_STACK_USAGE
875 static void check_stack_usage(void)
877 static DEFINE_SPINLOCK(low_water_lock);
878 static int lowest_to_date = THREAD_SIZE;
881 free = stack_not_used(current);
883 if (free >= lowest_to_date)
886 spin_lock(&low_water_lock);
887 if (free < lowest_to_date) {
888 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
890 current->comm, free);
891 lowest_to_date = free;
893 spin_unlock(&low_water_lock);
896 static inline void check_stack_usage(void) {}
899 NORET_TYPE void do_exit(long code)
901 struct task_struct *tsk = current;
904 profile_task_exit(tsk);
906 WARN_ON(atomic_read(&tsk->fs_excl));
908 if (unlikely(in_interrupt()))
909 panic("Aiee, killing interrupt handler!");
910 if (unlikely(!tsk->pid))
911 panic("Attempted to kill the idle task!");
913 tracehook_report_exit(&code);
915 validate_creds_for_do_exit(tsk);
918 * We're taking recursive faults here in do_exit. Safest is to just
919 * leave this task alone and wait for reboot.
921 if (unlikely(tsk->flags & PF_EXITING)) {
923 "Fixing recursive fault but reboot is needed!\n");
925 * We can do this unlocked here. The futex code uses
926 * this flag just to verify whether the pi state
927 * cleanup has been done or not. In the worst case it
928 * loops once more. We pretend that the cleanup was
929 * done as there is no way to return. Either the
930 * OWNER_DIED bit is set by now or we push the blocked
931 * task into the wait for ever nirwana as well.
933 tsk->flags |= PF_EXITPIDONE;
934 set_current_state(TASK_UNINTERRUPTIBLE);
940 exit_signals(tsk); /* sets PF_EXITING */
942 * tsk->flags are checked in the futex code to protect against
943 * an exiting task cleaning up the robust pi futexes.
946 raw_spin_unlock_wait(&tsk->pi_lock);
948 if (unlikely(in_atomic()))
949 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
950 current->comm, task_pid_nr(current),
953 acct_update_integrals(tsk);
954 /* sync mm's RSS info before statistics gathering */
956 sync_mm_rss(tsk, tsk->mm);
957 group_dead = atomic_dec_and_test(&tsk->signal->live);
959 hrtimer_cancel(&tsk->signal->real_timer);
960 exit_itimers(tsk->signal);
962 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
964 acct_collect(code, group_dead);
967 if (unlikely(tsk->audit_context))
970 tsk->exit_code = code;
971 taskstats_exit(tsk, group_dead);
977 trace_sched_process_exit(tsk);
987 disassociate_ctty(1);
989 module_put(task_thread_info(tsk)->exec_domain->module);
991 proc_exit_connector(tsk);
994 * FIXME: do that only when needed, using sched_exit tracepoint
996 flush_ptrace_hw_breakpoint(tsk);
998 * Flush inherited counters to the parent - before the parent
999 * gets woken up by child-exit notifications.
1001 perf_event_exit_task(tsk);
1003 exit_notify(tsk, group_dead);
1006 mpol_put(tsk->mempolicy);
1007 tsk->mempolicy = NULL;
1011 if (unlikely(current->pi_state_cache))
1012 kfree(current->pi_state_cache);
1015 * Make sure we are holding no locks:
1017 debug_check_no_locks_held(tsk);
1019 * We can do this unlocked here. The futex code uses this flag
1020 * just to verify whether the pi state cleanup has been done
1021 * or not. In the worst case it loops once more.
1023 tsk->flags |= PF_EXITPIDONE;
1025 if (tsk->io_context)
1026 exit_io_context(tsk);
1028 if (tsk->splice_pipe)
1029 __free_pipe_info(tsk->splice_pipe);
1031 validate_creds_for_do_exit(tsk);
1035 /* causes final put_task_struct in finish_task_switch(). */
1036 tsk->state = TASK_DEAD;
1039 /* Avoid "noreturn function does return". */
1041 cpu_relax(); /* For when BUG is null */
1044 EXPORT_SYMBOL_GPL(do_exit);
1046 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1054 EXPORT_SYMBOL(complete_and_exit);
1056 SYSCALL_DEFINE1(exit, int, error_code)
1058 do_exit((error_code&0xff)<<8);
1062 * Take down every thread in the group. This is called by fatal signals
1063 * as well as by sys_exit_group (below).
1066 do_group_exit(int exit_code)
1068 struct signal_struct *sig = current->signal;
1070 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1072 if (signal_group_exit(sig))
1073 exit_code = sig->group_exit_code;
1074 else if (!thread_group_empty(current)) {
1075 struct sighand_struct *const sighand = current->sighand;
1076 spin_lock_irq(&sighand->siglock);
1077 if (signal_group_exit(sig))
1078 /* Another thread got here before we took the lock. */
1079 exit_code = sig->group_exit_code;
1081 sig->group_exit_code = exit_code;
1082 sig->flags = SIGNAL_GROUP_EXIT;
1083 zap_other_threads(current);
1085 spin_unlock_irq(&sighand->siglock);
1093 * this kills every thread in the thread group. Note that any externally
1094 * wait4()-ing process will get the correct exit code - even if this
1095 * thread is not the thread group leader.
1097 SYSCALL_DEFINE1(exit_group, int, error_code)
1099 do_group_exit((error_code & 0xff) << 8);
1105 enum pid_type wo_type;
1109 struct siginfo __user *wo_info;
1110 int __user *wo_stat;
1111 struct rusage __user *wo_rusage;
1113 wait_queue_t child_wait;
1118 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1120 if (type != PIDTYPE_PID)
1121 task = task->group_leader;
1122 return task->pids[type].pid;
1125 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1127 return wo->wo_type == PIDTYPE_MAX ||
1128 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1131 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1133 if (!eligible_pid(wo, p))
1135 /* Wait for all children (clone and not) if __WALL is set;
1136 * otherwise, wait for clone children *only* if __WCLONE is
1137 * set; otherwise, wait for non-clone children *only*. (Note:
1138 * A "clone" child here is one that reports to its parent
1139 * using a signal other than SIGCHLD.) */
1140 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1141 && !(wo->wo_flags & __WALL))
1147 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1148 pid_t pid, uid_t uid, int why, int status)
1150 struct siginfo __user *infop;
1151 int retval = wo->wo_rusage
1152 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1155 infop = wo->wo_info;
1158 retval = put_user(SIGCHLD, &infop->si_signo);
1160 retval = put_user(0, &infop->si_errno);
1162 retval = put_user((short)why, &infop->si_code);
1164 retval = put_user(pid, &infop->si_pid);
1166 retval = put_user(uid, &infop->si_uid);
1168 retval = put_user(status, &infop->si_status);
1176 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1177 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1178 * the lock and this task is uninteresting. If we return nonzero, we have
1179 * released the lock and the system call should return.
1181 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1183 unsigned long state;
1184 int retval, status, traced;
1185 pid_t pid = task_pid_vnr(p);
1186 uid_t uid = __task_cred(p)->uid;
1187 struct siginfo __user *infop;
1189 if (!likely(wo->wo_flags & WEXITED))
1192 if (unlikely(wo->wo_flags & WNOWAIT)) {
1193 int exit_code = p->exit_code;
1197 read_unlock(&tasklist_lock);
1198 if ((exit_code & 0x7f) == 0) {
1200 status = exit_code >> 8;
1202 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1203 status = exit_code & 0x7f;
1205 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1209 * Try to move the task's state to DEAD
1210 * only one thread is allowed to do this:
1212 state = xchg(&p->exit_state, EXIT_DEAD);
1213 if (state != EXIT_ZOMBIE) {
1214 BUG_ON(state != EXIT_DEAD);
1218 traced = ptrace_reparented(p);
1220 * It can be ptraced but not reparented, check
1221 * !task_detached() to filter out sub-threads.
1223 if (likely(!traced) && likely(!task_detached(p))) {
1224 struct signal_struct *psig;
1225 struct signal_struct *sig;
1226 unsigned long maxrss;
1227 cputime_t tgutime, tgstime;
1230 * The resource counters for the group leader are in its
1231 * own task_struct. Those for dead threads in the group
1232 * are in its signal_struct, as are those for the child
1233 * processes it has previously reaped. All these
1234 * accumulate in the parent's signal_struct c* fields.
1236 * We don't bother to take a lock here to protect these
1237 * p->signal fields, because they are only touched by
1238 * __exit_signal, which runs with tasklist_lock
1239 * write-locked anyway, and so is excluded here. We do
1240 * need to protect the access to parent->signal fields,
1241 * as other threads in the parent group can be right
1242 * here reaping other children at the same time.
1244 * We use thread_group_times() to get times for the thread
1245 * group, which consolidates times for all threads in the
1246 * group including the group leader.
1248 thread_group_times(p, &tgutime, &tgstime);
1249 spin_lock_irq(&p->real_parent->sighand->siglock);
1250 psig = p->real_parent->signal;
1253 cputime_add(psig->cutime,
1254 cputime_add(tgutime,
1257 cputime_add(psig->cstime,
1258 cputime_add(tgstime,
1261 cputime_add(psig->cgtime,
1262 cputime_add(p->gtime,
1263 cputime_add(sig->gtime,
1266 p->min_flt + sig->min_flt + sig->cmin_flt;
1268 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1270 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1272 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1274 task_io_get_inblock(p) +
1275 sig->inblock + sig->cinblock;
1277 task_io_get_oublock(p) +
1278 sig->oublock + sig->coublock;
1279 maxrss = max(sig->maxrss, sig->cmaxrss);
1280 if (psig->cmaxrss < maxrss)
1281 psig->cmaxrss = maxrss;
1282 task_io_accounting_add(&psig->ioac, &p->ioac);
1283 task_io_accounting_add(&psig->ioac, &sig->ioac);
1284 spin_unlock_irq(&p->real_parent->sighand->siglock);
1288 * Now we are sure this task is interesting, and no other
1289 * thread can reap it because we set its state to EXIT_DEAD.
1291 read_unlock(&tasklist_lock);
1293 retval = wo->wo_rusage
1294 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1295 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1296 ? p->signal->group_exit_code : p->exit_code;
1297 if (!retval && wo->wo_stat)
1298 retval = put_user(status, wo->wo_stat);
1300 infop = wo->wo_info;
1301 if (!retval && infop)
1302 retval = put_user(SIGCHLD, &infop->si_signo);
1303 if (!retval && infop)
1304 retval = put_user(0, &infop->si_errno);
1305 if (!retval && infop) {
1308 if ((status & 0x7f) == 0) {
1312 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1315 retval = put_user((short)why, &infop->si_code);
1317 retval = put_user(status, &infop->si_status);
1319 if (!retval && infop)
1320 retval = put_user(pid, &infop->si_pid);
1321 if (!retval && infop)
1322 retval = put_user(uid, &infop->si_uid);
1327 write_lock_irq(&tasklist_lock);
1328 /* We dropped tasklist, ptracer could die and untrace */
1331 * If this is not a detached task, notify the parent.
1332 * If it's still not detached after that, don't release
1335 if (!task_detached(p)) {
1336 do_notify_parent(p, p->exit_signal);
1337 if (!task_detached(p)) {
1338 p->exit_state = EXIT_ZOMBIE;
1342 write_unlock_irq(&tasklist_lock);
1350 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1353 if (task_is_stopped_or_traced(p))
1354 return &p->exit_code;
1356 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1357 return &p->signal->group_exit_code;
1363 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1364 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1365 * the lock and this task is uninteresting. If we return nonzero, we have
1366 * released the lock and the system call should return.
1368 static int wait_task_stopped(struct wait_opts *wo,
1369 int ptrace, struct task_struct *p)
1371 struct siginfo __user *infop;
1372 int retval, exit_code, *p_code, why;
1373 uid_t uid = 0; /* unneeded, required by compiler */
1377 * Traditionally we see ptrace'd stopped tasks regardless of options.
1379 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1383 spin_lock_irq(&p->sighand->siglock);
1385 p_code = task_stopped_code(p, ptrace);
1386 if (unlikely(!p_code))
1389 exit_code = *p_code;
1393 if (!unlikely(wo->wo_flags & WNOWAIT))
1396 /* don't need the RCU readlock here as we're holding a spinlock */
1397 uid = __task_cred(p)->uid;
1399 spin_unlock_irq(&p->sighand->siglock);
1404 * Now we are pretty sure this task is interesting.
1405 * Make sure it doesn't get reaped out from under us while we
1406 * give up the lock and then examine it below. We don't want to
1407 * keep holding onto the tasklist_lock while we call getrusage and
1408 * possibly take page faults for user memory.
1411 pid = task_pid_vnr(p);
1412 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1413 read_unlock(&tasklist_lock);
1415 if (unlikely(wo->wo_flags & WNOWAIT))
1416 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1418 retval = wo->wo_rusage
1419 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1420 if (!retval && wo->wo_stat)
1421 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1423 infop = wo->wo_info;
1424 if (!retval && infop)
1425 retval = put_user(SIGCHLD, &infop->si_signo);
1426 if (!retval && infop)
1427 retval = put_user(0, &infop->si_errno);
1428 if (!retval && infop)
1429 retval = put_user((short)why, &infop->si_code);
1430 if (!retval && infop)
1431 retval = put_user(exit_code, &infop->si_status);
1432 if (!retval && infop)
1433 retval = put_user(pid, &infop->si_pid);
1434 if (!retval && infop)
1435 retval = put_user(uid, &infop->si_uid);
1445 * Handle do_wait work for one task in a live, non-stopped state.
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_continued(struct wait_opts *wo, struct task_struct *p)
1456 if (!unlikely(wo->wo_flags & WCONTINUED))
1459 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1462 spin_lock_irq(&p->sighand->siglock);
1463 /* Re-check with the lock held. */
1464 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1465 spin_unlock_irq(&p->sighand->siglock);
1468 if (!unlikely(wo->wo_flags & WNOWAIT))
1469 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1470 uid = __task_cred(p)->uid;
1471 spin_unlock_irq(&p->sighand->siglock);
1473 pid = task_pid_vnr(p);
1475 read_unlock(&tasklist_lock);
1478 retval = wo->wo_rusage
1479 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1481 if (!retval && wo->wo_stat)
1482 retval = put_user(0xffff, wo->wo_stat);
1486 retval = wait_noreap_copyout(wo, p, pid, uid,
1487 CLD_CONTINUED, SIGCONT);
1488 BUG_ON(retval == 0);
1495 * Consider @p for a wait by @parent.
1497 * -ECHILD should be in ->notask_error before the first call.
1498 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1499 * Returns zero if the search for a child should continue;
1500 * then ->notask_error is 0 if @p is an eligible child,
1501 * or another error from security_task_wait(), or still -ECHILD.
1503 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1504 struct task_struct *p)
1506 int ret = eligible_child(wo, p);
1510 ret = security_task_wait(p);
1511 if (unlikely(ret < 0)) {
1513 * If we have not yet seen any eligible child,
1514 * then let this error code replace -ECHILD.
1515 * A permission error will give the user a clue
1516 * to look for security policy problems, rather
1517 * than for mysterious wait bugs.
1519 if (wo->notask_error)
1520 wo->notask_error = ret;
1524 if (likely(!ptrace) && unlikely(task_ptrace(p))) {
1526 * This child is hidden by ptrace.
1527 * We aren't allowed to see it now, but eventually we will.
1529 wo->notask_error = 0;
1533 if (p->exit_state == EXIT_DEAD)
1537 * We don't reap group leaders with subthreads.
1539 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1540 return wait_task_zombie(wo, p);
1543 * It's stopped or running now, so it might
1544 * later continue, exit, or stop again.
1546 wo->notask_error = 0;
1548 if (task_stopped_code(p, ptrace))
1549 return wait_task_stopped(wo, ptrace, p);
1551 return wait_task_continued(wo, p);
1555 * Do the work of do_wait() for one thread in the group, @tsk.
1557 * -ECHILD should be in ->notask_error before the first call.
1558 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1559 * Returns zero if the search for a child should continue; then
1560 * ->notask_error is 0 if there were any eligible children,
1561 * or another error from security_task_wait(), or still -ECHILD.
1563 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1565 struct task_struct *p;
1567 list_for_each_entry(p, &tsk->children, sibling) {
1568 int ret = wait_consider_task(wo, 0, p);
1576 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1578 struct task_struct *p;
1580 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1581 int ret = wait_consider_task(wo, 1, p);
1589 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1590 int sync, void *key)
1592 struct wait_opts *wo = container_of(wait, struct wait_opts,
1594 struct task_struct *p = key;
1596 if (!eligible_pid(wo, p))
1599 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1602 return default_wake_function(wait, mode, sync, key);
1605 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1607 __wake_up_sync_key(&parent->signal->wait_chldexit,
1608 TASK_INTERRUPTIBLE, 1, p);
1611 static long do_wait(struct wait_opts *wo)
1613 struct task_struct *tsk;
1616 trace_sched_process_wait(wo->wo_pid);
1618 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1619 wo->child_wait.private = current;
1620 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1623 * If there is nothing that can match our critiera just get out.
1624 * We will clear ->notask_error to zero if we see any child that
1625 * might later match our criteria, even if we are not able to reap
1628 wo->notask_error = -ECHILD;
1629 if ((wo->wo_type < PIDTYPE_MAX) &&
1630 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1633 set_current_state(TASK_INTERRUPTIBLE);
1634 read_lock(&tasklist_lock);
1637 retval = do_wait_thread(wo, tsk);
1641 retval = ptrace_do_wait(wo, tsk);
1645 if (wo->wo_flags & __WNOTHREAD)
1647 } while_each_thread(current, tsk);
1648 read_unlock(&tasklist_lock);
1651 retval = wo->notask_error;
1652 if (!retval && !(wo->wo_flags & WNOHANG)) {
1653 retval = -ERESTARTSYS;
1654 if (!signal_pending(current)) {
1660 __set_current_state(TASK_RUNNING);
1661 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1665 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1666 infop, int, options, struct rusage __user *, ru)
1668 struct wait_opts wo;
1669 struct pid *pid = NULL;
1673 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1675 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1688 type = PIDTYPE_PGID;
1696 if (type < PIDTYPE_MAX)
1697 pid = find_get_pid(upid);
1701 wo.wo_flags = options;
1711 * For a WNOHANG return, clear out all the fields
1712 * we would set so the user can easily tell the
1716 ret = put_user(0, &infop->si_signo);
1718 ret = put_user(0, &infop->si_errno);
1720 ret = put_user(0, &infop->si_code);
1722 ret = put_user(0, &infop->si_pid);
1724 ret = put_user(0, &infop->si_uid);
1726 ret = put_user(0, &infop->si_status);
1731 /* avoid REGPARM breakage on x86: */
1732 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1736 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1737 int, options, struct rusage __user *, ru)
1739 struct wait_opts wo;
1740 struct pid *pid = NULL;
1744 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1745 __WNOTHREAD|__WCLONE|__WALL))
1750 else if (upid < 0) {
1751 type = PIDTYPE_PGID;
1752 pid = find_get_pid(-upid);
1753 } else if (upid == 0) {
1754 type = PIDTYPE_PGID;
1755 pid = get_task_pid(current, PIDTYPE_PGID);
1756 } else /* upid > 0 */ {
1758 pid = find_get_pid(upid);
1763 wo.wo_flags = options | WEXITED;
1765 wo.wo_stat = stat_addr;
1770 /* avoid REGPARM breakage on x86: */
1771 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1775 #ifdef __ARCH_WANT_SYS_WAITPID
1778 * sys_waitpid() remains for compatibility. waitpid() should be
1779 * implemented by calling sys_wait4() from libc.a.
1781 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1783 return sys_wait4(pid, stat_addr, options, NULL);