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;
83 struct tty_struct *uninitialized_var(tty);
86 BUG_ON(!atomic_read(&sig->count));
88 sighand = rcu_dereference_check(tsk->sighand,
89 rcu_read_lock_held() ||
90 lockdep_tasklist_lock_is_held());
91 spin_lock(&sighand->siglock);
92 atomic_dec(&sig->count);
94 posix_cpu_timers_exit(tsk);
95 if (thread_group_leader(tsk)) {
96 posix_cpu_timers_exit_group(tsk);
101 * If there is any task waiting for the group exit
104 if (sig->notify_count > 0 && !--sig->notify_count)
105 wake_up_process(sig->group_exit_task);
107 if (tsk == sig->curr_target)
108 sig->curr_target = next_thread(tsk);
110 * Accumulate here the counters for all threads but the
111 * group leader as they die, so they can be added into
112 * the process-wide totals when those are taken.
113 * The group leader stays around as a zombie as long
114 * as there are other threads. When it gets reaped,
115 * the exit.c code will add its counts into these totals.
116 * We won't ever get here for the group leader, since it
117 * will have been the last reference on the signal_struct.
119 sig->utime = cputime_add(sig->utime, tsk->utime);
120 sig->stime = cputime_add(sig->stime, tsk->stime);
121 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
122 sig->min_flt += tsk->min_flt;
123 sig->maj_flt += tsk->maj_flt;
124 sig->nvcsw += tsk->nvcsw;
125 sig->nivcsw += tsk->nivcsw;
126 sig->inblock += task_io_get_inblock(tsk);
127 sig->oublock += task_io_get_oublock(tsk);
128 task_io_accounting_add(&sig->ioac, &tsk->ioac);
129 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
130 sig = NULL; /* Marker for below. */
133 __unhash_process(tsk);
136 * Do this under ->siglock, we can race with another thread
137 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
139 flush_sigqueue(&tsk->pending);
141 spin_unlock(&sighand->siglock);
143 __cleanup_sighand(sighand);
144 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
146 flush_sigqueue(&sig->shared_pending);
147 taskstats_tgid_free(sig);
149 * Make sure ->signal can't go away under rq->lock,
150 * see account_group_exec_runtime().
152 task_rq_unlock_wait(tsk);
157 static void delayed_put_task_struct(struct rcu_head *rhp)
159 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161 #ifdef CONFIG_PERF_EVENTS
162 WARN_ON_ONCE(tsk->perf_event_ctxp);
164 trace_sched_process_free(tsk);
165 put_task_struct(tsk);
169 void release_task(struct task_struct * p)
171 struct task_struct *leader;
174 tracehook_prepare_release_task(p);
175 /* don't need to get the RCU readlock here - the process is dead and
176 * can't be modifying its own credentials. But shut RCU-lockdep up */
178 atomic_dec(&__task_cred(p)->user->processes);
183 write_lock_irq(&tasklist_lock);
184 tracehook_finish_release_task(p);
188 * If we are the last non-leader member of the thread
189 * group, and the leader is zombie, then notify the
190 * group leader's parent process. (if it wants notification.)
193 leader = p->group_leader;
194 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
195 BUG_ON(task_detached(leader));
196 do_notify_parent(leader, leader->exit_signal);
198 * If we were the last child thread and the leader has
199 * exited already, and the leader's parent ignores SIGCHLD,
200 * then we are the one who should release the leader.
202 * do_notify_parent() will have marked it self-reaping in
205 zap_leader = task_detached(leader);
208 * This maintains the invariant that release_task()
209 * only runs on a task in EXIT_DEAD, just for sanity.
212 leader->exit_state = EXIT_DEAD;
215 write_unlock_irq(&tasklist_lock);
217 call_rcu(&p->rcu, delayed_put_task_struct);
220 if (unlikely(zap_leader))
225 * This checks not only the pgrp, but falls back on the pid if no
226 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
229 * The caller must hold rcu lock or the tasklist lock.
231 struct pid *session_of_pgrp(struct pid *pgrp)
233 struct task_struct *p;
234 struct pid *sid = NULL;
236 p = pid_task(pgrp, PIDTYPE_PGID);
238 p = pid_task(pgrp, PIDTYPE_PID);
240 sid = task_session(p);
246 * Determine if a process group is "orphaned", according to the POSIX
247 * definition in 2.2.2.52. Orphaned process groups are not to be affected
248 * by terminal-generated stop signals. Newly orphaned process groups are
249 * to receive a SIGHUP and a SIGCONT.
251 * "I ask you, have you ever known what it is to be an orphan?"
253 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
255 struct task_struct *p;
257 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
258 if ((p == ignored_task) ||
259 (p->exit_state && thread_group_empty(p)) ||
260 is_global_init(p->real_parent))
263 if (task_pgrp(p->real_parent) != pgrp &&
264 task_session(p->real_parent) == task_session(p))
266 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
271 int is_current_pgrp_orphaned(void)
275 read_lock(&tasklist_lock);
276 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
277 read_unlock(&tasklist_lock);
282 static int has_stopped_jobs(struct pid *pgrp)
285 struct task_struct *p;
287 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
288 if (!task_is_stopped(p))
292 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
297 * Check to see if any process groups have become orphaned as
298 * a result of our exiting, and if they have any stopped jobs,
299 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
302 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
304 struct pid *pgrp = task_pgrp(tsk);
305 struct task_struct *ignored_task = tsk;
308 /* exit: our father is in a different pgrp than
309 * we are and we were the only connection outside.
311 parent = tsk->real_parent;
313 /* reparent: our child is in a different pgrp than
314 * we are, and it was the only connection outside.
318 if (task_pgrp(parent) != pgrp &&
319 task_session(parent) == task_session(tsk) &&
320 will_become_orphaned_pgrp(pgrp, ignored_task) &&
321 has_stopped_jobs(pgrp)) {
322 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
323 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
328 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
330 * If a kernel thread is launched as a result of a system call, or if
331 * it ever exits, it should generally reparent itself to kthreadd so it
332 * isn't in the way of other processes and is correctly cleaned up on exit.
334 * The various task state such as scheduling policy and priority may have
335 * been inherited from a user process, so we reset them to sane values here.
337 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
339 static void reparent_to_kthreadd(void)
341 write_lock_irq(&tasklist_lock);
343 ptrace_unlink(current);
344 /* Reparent to init */
345 current->real_parent = current->parent = kthreadd_task;
346 list_move_tail(¤t->sibling, ¤t->real_parent->children);
348 /* Set the exit signal to SIGCHLD so we signal init on exit */
349 current->exit_signal = SIGCHLD;
351 if (task_nice(current) < 0)
352 set_user_nice(current, 0);
356 memcpy(current->signal->rlim, init_task.signal->rlim,
357 sizeof(current->signal->rlim));
359 atomic_inc(&init_cred.usage);
360 commit_creds(&init_cred);
361 write_unlock_irq(&tasklist_lock);
364 void __set_special_pids(struct pid *pid)
366 struct task_struct *curr = current->group_leader;
368 if (task_session(curr) != pid)
369 change_pid(curr, PIDTYPE_SID, pid);
371 if (task_pgrp(curr) != pid)
372 change_pid(curr, PIDTYPE_PGID, pid);
375 static void set_special_pids(struct pid *pid)
377 write_lock_irq(&tasklist_lock);
378 __set_special_pids(pid);
379 write_unlock_irq(&tasklist_lock);
383 * Let kernel threads use this to say that they allow a certain signal.
384 * Must not be used if kthread was cloned with CLONE_SIGHAND.
386 int allow_signal(int sig)
388 if (!valid_signal(sig) || sig < 1)
391 spin_lock_irq(¤t->sighand->siglock);
392 /* This is only needed for daemonize()'ed kthreads */
393 sigdelset(¤t->blocked, sig);
395 * Kernel threads handle their own signals. Let the signal code
396 * know it'll be handled, so that they don't get converted to
397 * SIGKILL or just silently dropped.
399 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
401 spin_unlock_irq(¤t->sighand->siglock);
405 EXPORT_SYMBOL(allow_signal);
407 int disallow_signal(int sig)
409 if (!valid_signal(sig) || sig < 1)
412 spin_lock_irq(¤t->sighand->siglock);
413 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
415 spin_unlock_irq(¤t->sighand->siglock);
419 EXPORT_SYMBOL(disallow_signal);
422 * Put all the gunge required to become a kernel thread without
423 * attached user resources in one place where it belongs.
426 void daemonize(const char *name, ...)
431 va_start(args, name);
432 vsnprintf(current->comm, sizeof(current->comm), name, args);
436 * If we were started as result of loading a module, close all of the
437 * user space pages. We don't need them, and if we didn't close them
438 * they would be locked into memory.
442 * We don't want to have TIF_FREEZE set if the system-wide hibernation
443 * or suspend transition begins right now.
445 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
447 if (current->nsproxy != &init_nsproxy) {
448 get_nsproxy(&init_nsproxy);
449 switch_task_namespaces(current, &init_nsproxy);
451 set_special_pids(&init_struct_pid);
452 proc_clear_tty(current);
454 /* Block and flush all signals */
455 sigfillset(&blocked);
456 sigprocmask(SIG_BLOCK, &blocked, NULL);
457 flush_signals(current);
459 /* Become as one with the init task */
461 daemonize_fs_struct();
463 current->files = init_task.files;
464 atomic_inc(¤t->files->count);
466 reparent_to_kthreadd();
469 EXPORT_SYMBOL(daemonize);
471 static void close_files(struct files_struct * files)
479 * It is safe to dereference the fd table without RCU or
480 * ->file_lock because this is the last reference to the
481 * files structure. But use RCU to shut RCU-lockdep up.
484 fdt = files_fdtable(files);
489 if (i >= fdt->max_fds)
491 set = fdt->open_fds->fds_bits[j++];
494 struct file * file = xchg(&fdt->fd[i], NULL);
496 filp_close(file, files);
506 struct files_struct *get_files_struct(struct task_struct *task)
508 struct files_struct *files;
513 atomic_inc(&files->count);
519 void put_files_struct(struct files_struct *files)
523 if (atomic_dec_and_test(&files->count)) {
526 * Free the fd and fdset arrays if we expanded them.
527 * If the fdtable was embedded, pass files for freeing
528 * at the end of the RCU grace period. Otherwise,
529 * you can free files immediately.
532 fdt = files_fdtable(files);
533 if (fdt != &files->fdtab)
534 kmem_cache_free(files_cachep, files);
540 void reset_files_struct(struct files_struct *files)
542 struct task_struct *tsk = current;
543 struct files_struct *old;
549 put_files_struct(old);
552 void exit_files(struct task_struct *tsk)
554 struct files_struct * files = tsk->files;
560 put_files_struct(files);
564 #ifdef CONFIG_MM_OWNER
566 * Task p is exiting and it owned mm, lets find a new owner for it
569 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
572 * If there are other users of the mm and the owner (us) is exiting
573 * we need to find a new owner to take on the responsibility.
575 if (atomic_read(&mm->mm_users) <= 1)
582 void mm_update_next_owner(struct mm_struct *mm)
584 struct task_struct *c, *g, *p = current;
587 if (!mm_need_new_owner(mm, p))
590 read_lock(&tasklist_lock);
592 * Search in the children
594 list_for_each_entry(c, &p->children, sibling) {
596 goto assign_new_owner;
600 * Search in the siblings
602 list_for_each_entry(c, &p->real_parent->children, sibling) {
604 goto assign_new_owner;
608 * Search through everything else. We should not get
611 do_each_thread(g, c) {
613 goto assign_new_owner;
614 } while_each_thread(g, c);
616 read_unlock(&tasklist_lock);
618 * We found no owner yet mm_users > 1: this implies that we are
619 * most likely racing with swapoff (try_to_unuse()) or /proc or
620 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
629 * The task_lock protects c->mm from changing.
630 * We always want mm->owner->mm == mm
634 * Delay read_unlock() till we have the task_lock()
635 * to ensure that c does not slip away underneath us
637 read_unlock(&tasklist_lock);
647 #endif /* CONFIG_MM_OWNER */
650 * Turn us into a lazy TLB process if we
653 static void exit_mm(struct task_struct * tsk)
655 struct mm_struct *mm = tsk->mm;
656 struct core_state *core_state;
662 * Serialize with any possible pending coredump.
663 * We must hold mmap_sem around checking core_state
664 * and clearing tsk->mm. The core-inducing thread
665 * will increment ->nr_threads for each thread in the
666 * group with ->mm != NULL.
668 down_read(&mm->mmap_sem);
669 core_state = mm->core_state;
671 struct core_thread self;
672 up_read(&mm->mmap_sem);
675 self.next = xchg(&core_state->dumper.next, &self);
677 * Implies mb(), the result of xchg() must be visible
678 * to core_state->dumper.
680 if (atomic_dec_and_test(&core_state->nr_threads))
681 complete(&core_state->startup);
684 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
685 if (!self.task) /* see coredump_finish() */
689 __set_task_state(tsk, TASK_RUNNING);
690 down_read(&mm->mmap_sem);
692 atomic_inc(&mm->mm_count);
693 BUG_ON(mm != tsk->active_mm);
694 /* more a memory barrier than a real lock */
697 up_read(&mm->mmap_sem);
698 enter_lazy_tlb(mm, current);
699 /* We don't want this task to be frozen prematurely */
700 clear_freeze_flag(tsk);
702 mm_update_next_owner(mm);
707 * When we die, we re-parent all our children.
708 * Try to give them to another thread in our thread
709 * group, and if no such member exists, give it to
710 * the child reaper process (ie "init") in our pid
713 static struct task_struct *find_new_reaper(struct task_struct *father)
715 struct pid_namespace *pid_ns = task_active_pid_ns(father);
716 struct task_struct *thread;
719 while_each_thread(father, thread) {
720 if (thread->flags & PF_EXITING)
722 if (unlikely(pid_ns->child_reaper == father))
723 pid_ns->child_reaper = thread;
727 if (unlikely(pid_ns->child_reaper == father)) {
728 write_unlock_irq(&tasklist_lock);
729 if (unlikely(pid_ns == &init_pid_ns))
730 panic("Attempted to kill init!");
732 zap_pid_ns_processes(pid_ns);
733 write_lock_irq(&tasklist_lock);
735 * We can not clear ->child_reaper or leave it alone.
736 * There may by stealth EXIT_DEAD tasks on ->children,
737 * forget_original_parent() must move them somewhere.
739 pid_ns->child_reaper = init_pid_ns.child_reaper;
742 return pid_ns->child_reaper;
746 * Any that need to be release_task'd are put on the @dead list.
748 static void reparent_leader(struct task_struct *father, struct task_struct *p,
749 struct list_head *dead)
751 list_move_tail(&p->sibling, &p->real_parent->children);
753 if (task_detached(p))
756 * If this is a threaded reparent there is no need to
757 * notify anyone anything has happened.
759 if (same_thread_group(p->real_parent, father))
762 /* We don't want people slaying init. */
763 p->exit_signal = SIGCHLD;
765 /* If it has exited notify the new parent about this child's death. */
766 if (!task_ptrace(p) &&
767 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
768 do_notify_parent(p, p->exit_signal);
769 if (task_detached(p)) {
770 p->exit_state = EXIT_DEAD;
771 list_move_tail(&p->sibling, dead);
775 kill_orphaned_pgrp(p, father);
778 static void forget_original_parent(struct task_struct *father)
780 struct task_struct *p, *n, *reaper;
781 LIST_HEAD(dead_children);
785 write_lock_irq(&tasklist_lock);
786 reaper = find_new_reaper(father);
788 list_for_each_entry_safe(p, n, &father->children, sibling) {
789 struct task_struct *t = p;
791 t->real_parent = reaper;
792 if (t->parent == father) {
793 BUG_ON(task_ptrace(t));
794 t->parent = t->real_parent;
796 if (t->pdeath_signal)
797 group_send_sig_info(t->pdeath_signal,
799 } while_each_thread(p, t);
800 reparent_leader(father, p, &dead_children);
802 write_unlock_irq(&tasklist_lock);
804 BUG_ON(!list_empty(&father->children));
806 list_for_each_entry_safe(p, n, &dead_children, sibling) {
807 list_del_init(&p->sibling);
813 * Send signals to all our closest relatives so that they know
814 * to properly mourn us..
816 static void exit_notify(struct task_struct *tsk, int group_dead)
822 * This does two things:
824 * A. Make init inherit all the child processes
825 * B. Check to see if any process groups have become orphaned
826 * as a result of our exiting, and if they have any stopped
827 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
829 forget_original_parent(tsk);
830 exit_task_namespaces(tsk);
832 write_lock_irq(&tasklist_lock);
834 kill_orphaned_pgrp(tsk->group_leader, NULL);
836 /* Let father know we died
838 * Thread signals are configurable, but you aren't going to use
839 * that to send signals to arbitary processes.
840 * That stops right now.
842 * If the parent exec id doesn't match the exec id we saved
843 * when we started then we know the parent has changed security
846 * If our self_exec id doesn't match our parent_exec_id then
847 * we have changed execution domain as these two values started
848 * the same after a fork.
850 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
851 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
852 tsk->self_exec_id != tsk->parent_exec_id))
853 tsk->exit_signal = SIGCHLD;
855 signal = tracehook_notify_death(tsk, &cookie, group_dead);
857 signal = do_notify_parent(tsk, signal);
859 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
861 /* mt-exec, de_thread() is waiting for group leader */
862 if (unlikely(tsk->signal->notify_count < 0))
863 wake_up_process(tsk->signal->group_exit_task);
864 write_unlock_irq(&tasklist_lock);
866 tracehook_report_death(tsk, signal, cookie, group_dead);
868 /* If the process is dead, release it - nobody will wait for it */
869 if (signal == DEATH_REAP)
873 #ifdef CONFIG_DEBUG_STACK_USAGE
874 static void check_stack_usage(void)
876 static DEFINE_SPINLOCK(low_water_lock);
877 static int lowest_to_date = THREAD_SIZE;
880 free = stack_not_used(current);
882 if (free >= lowest_to_date)
885 spin_lock(&low_water_lock);
886 if (free < lowest_to_date) {
887 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
889 current->comm, free);
890 lowest_to_date = free;
892 spin_unlock(&low_water_lock);
895 static inline void check_stack_usage(void) {}
898 NORET_TYPE void do_exit(long code)
900 struct task_struct *tsk = current;
903 profile_task_exit(tsk);
905 WARN_ON(atomic_read(&tsk->fs_excl));
907 if (unlikely(in_interrupt()))
908 panic("Aiee, killing interrupt handler!");
909 if (unlikely(!tsk->pid))
910 panic("Attempted to kill the idle task!");
912 tracehook_report_exit(&code);
914 validate_creds_for_do_exit(tsk);
917 * We're taking recursive faults here in do_exit. Safest is to just
918 * leave this task alone and wait for reboot.
920 if (unlikely(tsk->flags & PF_EXITING)) {
922 "Fixing recursive fault but reboot is needed!\n");
924 * We can do this unlocked here. The futex code uses
925 * this flag just to verify whether the pi state
926 * cleanup has been done or not. In the worst case it
927 * loops once more. We pretend that the cleanup was
928 * done as there is no way to return. Either the
929 * OWNER_DIED bit is set by now or we push the blocked
930 * task into the wait for ever nirwana as well.
932 tsk->flags |= PF_EXITPIDONE;
933 set_current_state(TASK_UNINTERRUPTIBLE);
939 exit_signals(tsk); /* sets PF_EXITING */
941 * tsk->flags are checked in the futex code to protect against
942 * an exiting task cleaning up the robust pi futexes.
945 raw_spin_unlock_wait(&tsk->pi_lock);
947 if (unlikely(in_atomic()))
948 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
949 current->comm, task_pid_nr(current),
952 acct_update_integrals(tsk);
953 /* sync mm's RSS info before statistics gathering */
955 sync_mm_rss(tsk, tsk->mm);
956 group_dead = atomic_dec_and_test(&tsk->signal->live);
958 hrtimer_cancel(&tsk->signal->real_timer);
959 exit_itimers(tsk->signal);
961 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
963 acct_collect(code, group_dead);
966 if (unlikely(tsk->audit_context))
969 tsk->exit_code = code;
970 taskstats_exit(tsk, group_dead);
976 trace_sched_process_exit(tsk);
986 disassociate_ctty(1);
988 module_put(task_thread_info(tsk)->exec_domain->module);
990 proc_exit_connector(tsk);
993 * FIXME: do that only when needed, using sched_exit tracepoint
995 flush_ptrace_hw_breakpoint(tsk);
997 * Flush inherited counters to the parent - before the parent
998 * gets woken up by child-exit notifications.
1000 perf_event_exit_task(tsk);
1002 exit_notify(tsk, group_dead);
1005 mpol_put(tsk->mempolicy);
1006 tsk->mempolicy = NULL;
1010 if (unlikely(current->pi_state_cache))
1011 kfree(current->pi_state_cache);
1014 * Make sure we are holding no locks:
1016 debug_check_no_locks_held(tsk);
1018 * We can do this unlocked here. The futex code uses this flag
1019 * just to verify whether the pi state cleanup has been done
1020 * or not. In the worst case it loops once more.
1022 tsk->flags |= PF_EXITPIDONE;
1024 if (tsk->io_context)
1025 exit_io_context(tsk);
1027 if (tsk->splice_pipe)
1028 __free_pipe_info(tsk->splice_pipe);
1030 validate_creds_for_do_exit(tsk);
1034 /* causes final put_task_struct in finish_task_switch(). */
1035 tsk->state = TASK_DEAD;
1038 /* Avoid "noreturn function does return". */
1040 cpu_relax(); /* For when BUG is null */
1043 EXPORT_SYMBOL_GPL(do_exit);
1045 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1053 EXPORT_SYMBOL(complete_and_exit);
1055 SYSCALL_DEFINE1(exit, int, error_code)
1057 do_exit((error_code&0xff)<<8);
1061 * Take down every thread in the group. This is called by fatal signals
1062 * as well as by sys_exit_group (below).
1065 do_group_exit(int exit_code)
1067 struct signal_struct *sig = current->signal;
1069 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1071 if (signal_group_exit(sig))
1072 exit_code = sig->group_exit_code;
1073 else if (!thread_group_empty(current)) {
1074 struct sighand_struct *const sighand = current->sighand;
1075 spin_lock_irq(&sighand->siglock);
1076 if (signal_group_exit(sig))
1077 /* Another thread got here before we took the lock. */
1078 exit_code = sig->group_exit_code;
1080 sig->group_exit_code = exit_code;
1081 sig->flags = SIGNAL_GROUP_EXIT;
1082 zap_other_threads(current);
1084 spin_unlock_irq(&sighand->siglock);
1092 * this kills every thread in the thread group. Note that any externally
1093 * wait4()-ing process will get the correct exit code - even if this
1094 * thread is not the thread group leader.
1096 SYSCALL_DEFINE1(exit_group, int, error_code)
1098 do_group_exit((error_code & 0xff) << 8);
1104 enum pid_type wo_type;
1108 struct siginfo __user *wo_info;
1109 int __user *wo_stat;
1110 struct rusage __user *wo_rusage;
1112 wait_queue_t child_wait;
1117 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1119 if (type != PIDTYPE_PID)
1120 task = task->group_leader;
1121 return task->pids[type].pid;
1124 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1126 return wo->wo_type == PIDTYPE_MAX ||
1127 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1130 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1132 if (!eligible_pid(wo, p))
1134 /* Wait for all children (clone and not) if __WALL is set;
1135 * otherwise, wait for clone children *only* if __WCLONE is
1136 * set; otherwise, wait for non-clone children *only*. (Note:
1137 * A "clone" child here is one that reports to its parent
1138 * using a signal other than SIGCHLD.) */
1139 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1140 && !(wo->wo_flags & __WALL))
1146 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1147 pid_t pid, uid_t uid, int why, int status)
1149 struct siginfo __user *infop;
1150 int retval = wo->wo_rusage
1151 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1154 infop = wo->wo_info;
1157 retval = put_user(SIGCHLD, &infop->si_signo);
1159 retval = put_user(0, &infop->si_errno);
1161 retval = put_user((short)why, &infop->si_code);
1163 retval = put_user(pid, &infop->si_pid);
1165 retval = put_user(uid, &infop->si_uid);
1167 retval = put_user(status, &infop->si_status);
1175 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1176 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1177 * the lock and this task is uninteresting. If we return nonzero, we have
1178 * released the lock and the system call should return.
1180 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1182 unsigned long state;
1183 int retval, status, traced;
1184 pid_t pid = task_pid_vnr(p);
1185 uid_t uid = __task_cred(p)->uid;
1186 struct siginfo __user *infop;
1188 if (!likely(wo->wo_flags & WEXITED))
1191 if (unlikely(wo->wo_flags & WNOWAIT)) {
1192 int exit_code = p->exit_code;
1196 read_unlock(&tasklist_lock);
1197 if ((exit_code & 0x7f) == 0) {
1199 status = exit_code >> 8;
1201 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1202 status = exit_code & 0x7f;
1204 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1208 * Try to move the task's state to DEAD
1209 * only one thread is allowed to do this:
1211 state = xchg(&p->exit_state, EXIT_DEAD);
1212 if (state != EXIT_ZOMBIE) {
1213 BUG_ON(state != EXIT_DEAD);
1217 traced = ptrace_reparented(p);
1219 * It can be ptraced but not reparented, check
1220 * !task_detached() to filter out sub-threads.
1222 if (likely(!traced) && likely(!task_detached(p))) {
1223 struct signal_struct *psig;
1224 struct signal_struct *sig;
1225 unsigned long maxrss;
1226 cputime_t tgutime, tgstime;
1229 * The resource counters for the group leader are in its
1230 * own task_struct. Those for dead threads in the group
1231 * are in its signal_struct, as are those for the child
1232 * processes it has previously reaped. All these
1233 * accumulate in the parent's signal_struct c* fields.
1235 * We don't bother to take a lock here to protect these
1236 * p->signal fields, because they are only touched by
1237 * __exit_signal, which runs with tasklist_lock
1238 * write-locked anyway, and so is excluded here. We do
1239 * need to protect the access to parent->signal fields,
1240 * as other threads in the parent group can be right
1241 * here reaping other children at the same time.
1243 * We use thread_group_times() to get times for the thread
1244 * group, which consolidates times for all threads in the
1245 * group including the group leader.
1247 thread_group_times(p, &tgutime, &tgstime);
1248 spin_lock_irq(&p->real_parent->sighand->siglock);
1249 psig = p->real_parent->signal;
1252 cputime_add(psig->cutime,
1253 cputime_add(tgutime,
1256 cputime_add(psig->cstime,
1257 cputime_add(tgstime,
1260 cputime_add(psig->cgtime,
1261 cputime_add(p->gtime,
1262 cputime_add(sig->gtime,
1265 p->min_flt + sig->min_flt + sig->cmin_flt;
1267 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1269 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1271 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1273 task_io_get_inblock(p) +
1274 sig->inblock + sig->cinblock;
1276 task_io_get_oublock(p) +
1277 sig->oublock + sig->coublock;
1278 maxrss = max(sig->maxrss, sig->cmaxrss);
1279 if (psig->cmaxrss < maxrss)
1280 psig->cmaxrss = maxrss;
1281 task_io_accounting_add(&psig->ioac, &p->ioac);
1282 task_io_accounting_add(&psig->ioac, &sig->ioac);
1283 spin_unlock_irq(&p->real_parent->sighand->siglock);
1287 * Now we are sure this task is interesting, and no other
1288 * thread can reap it because we set its state to EXIT_DEAD.
1290 read_unlock(&tasklist_lock);
1292 retval = wo->wo_rusage
1293 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1294 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1295 ? p->signal->group_exit_code : p->exit_code;
1296 if (!retval && wo->wo_stat)
1297 retval = put_user(status, wo->wo_stat);
1299 infop = wo->wo_info;
1300 if (!retval && infop)
1301 retval = put_user(SIGCHLD, &infop->si_signo);
1302 if (!retval && infop)
1303 retval = put_user(0, &infop->si_errno);
1304 if (!retval && infop) {
1307 if ((status & 0x7f) == 0) {
1311 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1314 retval = put_user((short)why, &infop->si_code);
1316 retval = put_user(status, &infop->si_status);
1318 if (!retval && infop)
1319 retval = put_user(pid, &infop->si_pid);
1320 if (!retval && infop)
1321 retval = put_user(uid, &infop->si_uid);
1326 write_lock_irq(&tasklist_lock);
1327 /* We dropped tasklist, ptracer could die and untrace */
1330 * If this is not a detached task, notify the parent.
1331 * If it's still not detached after that, don't release
1334 if (!task_detached(p)) {
1335 do_notify_parent(p, p->exit_signal);
1336 if (!task_detached(p)) {
1337 p->exit_state = EXIT_ZOMBIE;
1341 write_unlock_irq(&tasklist_lock);
1349 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1352 if (task_is_stopped_or_traced(p))
1353 return &p->exit_code;
1355 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1356 return &p->signal->group_exit_code;
1362 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1363 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1364 * the lock and this task is uninteresting. If we return nonzero, we have
1365 * released the lock and the system call should return.
1367 static int wait_task_stopped(struct wait_opts *wo,
1368 int ptrace, struct task_struct *p)
1370 struct siginfo __user *infop;
1371 int retval, exit_code, *p_code, why;
1372 uid_t uid = 0; /* unneeded, required by compiler */
1376 * Traditionally we see ptrace'd stopped tasks regardless of options.
1378 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1382 spin_lock_irq(&p->sighand->siglock);
1384 p_code = task_stopped_code(p, ptrace);
1385 if (unlikely(!p_code))
1388 exit_code = *p_code;
1392 if (!unlikely(wo->wo_flags & WNOWAIT))
1395 /* don't need the RCU readlock here as we're holding a spinlock */
1396 uid = __task_cred(p)->uid;
1398 spin_unlock_irq(&p->sighand->siglock);
1403 * Now we are pretty sure this task is interesting.
1404 * Make sure it doesn't get reaped out from under us while we
1405 * give up the lock and then examine it below. We don't want to
1406 * keep holding onto the tasklist_lock while we call getrusage and
1407 * possibly take page faults for user memory.
1410 pid = task_pid_vnr(p);
1411 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1412 read_unlock(&tasklist_lock);
1414 if (unlikely(wo->wo_flags & WNOWAIT))
1415 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1417 retval = wo->wo_rusage
1418 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1419 if (!retval && wo->wo_stat)
1420 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1422 infop = wo->wo_info;
1423 if (!retval && infop)
1424 retval = put_user(SIGCHLD, &infop->si_signo);
1425 if (!retval && infop)
1426 retval = put_user(0, &infop->si_errno);
1427 if (!retval && infop)
1428 retval = put_user((short)why, &infop->si_code);
1429 if (!retval && infop)
1430 retval = put_user(exit_code, &infop->si_status);
1431 if (!retval && infop)
1432 retval = put_user(pid, &infop->si_pid);
1433 if (!retval && infop)
1434 retval = put_user(uid, &infop->si_uid);
1444 * Handle do_wait work for one task in a live, non-stopped state.
1445 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1446 * the lock and this task is uninteresting. If we return nonzero, we have
1447 * released the lock and the system call should return.
1449 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1455 if (!unlikely(wo->wo_flags & WCONTINUED))
1458 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1461 spin_lock_irq(&p->sighand->siglock);
1462 /* Re-check with the lock held. */
1463 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1464 spin_unlock_irq(&p->sighand->siglock);
1467 if (!unlikely(wo->wo_flags & WNOWAIT))
1468 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1469 uid = __task_cred(p)->uid;
1470 spin_unlock_irq(&p->sighand->siglock);
1472 pid = task_pid_vnr(p);
1474 read_unlock(&tasklist_lock);
1477 retval = wo->wo_rusage
1478 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1480 if (!retval && wo->wo_stat)
1481 retval = put_user(0xffff, wo->wo_stat);
1485 retval = wait_noreap_copyout(wo, p, pid, uid,
1486 CLD_CONTINUED, SIGCONT);
1487 BUG_ON(retval == 0);
1494 * Consider @p for a wait by @parent.
1496 * -ECHILD should be in ->notask_error before the first call.
1497 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1498 * Returns zero if the search for a child should continue;
1499 * then ->notask_error is 0 if @p is an eligible child,
1500 * or another error from security_task_wait(), or still -ECHILD.
1502 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1503 struct task_struct *p)
1505 int ret = eligible_child(wo, p);
1509 ret = security_task_wait(p);
1510 if (unlikely(ret < 0)) {
1512 * If we have not yet seen any eligible child,
1513 * then let this error code replace -ECHILD.
1514 * A permission error will give the user a clue
1515 * to look for security policy problems, rather
1516 * than for mysterious wait bugs.
1518 if (wo->notask_error)
1519 wo->notask_error = ret;
1523 if (likely(!ptrace) && unlikely(task_ptrace(p))) {
1525 * This child is hidden by ptrace.
1526 * We aren't allowed to see it now, but eventually we will.
1528 wo->notask_error = 0;
1532 if (p->exit_state == EXIT_DEAD)
1536 * We don't reap group leaders with subthreads.
1538 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1539 return wait_task_zombie(wo, p);
1542 * It's stopped or running now, so it might
1543 * later continue, exit, or stop again.
1545 wo->notask_error = 0;
1547 if (task_stopped_code(p, ptrace))
1548 return wait_task_stopped(wo, ptrace, p);
1550 return wait_task_continued(wo, p);
1554 * Do the work of do_wait() for one thread in the group, @tsk.
1556 * -ECHILD should be in ->notask_error before the first call.
1557 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1558 * Returns zero if the search for a child should continue; then
1559 * ->notask_error is 0 if there were any eligible children,
1560 * or another error from security_task_wait(), or still -ECHILD.
1562 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1564 struct task_struct *p;
1566 list_for_each_entry(p, &tsk->children, sibling) {
1567 int ret = wait_consider_task(wo, 0, p);
1575 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1577 struct task_struct *p;
1579 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1580 int ret = wait_consider_task(wo, 1, p);
1588 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1589 int sync, void *key)
1591 struct wait_opts *wo = container_of(wait, struct wait_opts,
1593 struct task_struct *p = key;
1595 if (!eligible_pid(wo, p))
1598 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1601 return default_wake_function(wait, mode, sync, key);
1604 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1606 __wake_up_sync_key(&parent->signal->wait_chldexit,
1607 TASK_INTERRUPTIBLE, 1, p);
1610 static long do_wait(struct wait_opts *wo)
1612 struct task_struct *tsk;
1615 trace_sched_process_wait(wo->wo_pid);
1617 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1618 wo->child_wait.private = current;
1619 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1622 * If there is nothing that can match our critiera just get out.
1623 * We will clear ->notask_error to zero if we see any child that
1624 * might later match our criteria, even if we are not able to reap
1627 wo->notask_error = -ECHILD;
1628 if ((wo->wo_type < PIDTYPE_MAX) &&
1629 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1632 set_current_state(TASK_INTERRUPTIBLE);
1633 read_lock(&tasklist_lock);
1636 retval = do_wait_thread(wo, tsk);
1640 retval = ptrace_do_wait(wo, tsk);
1644 if (wo->wo_flags & __WNOTHREAD)
1646 } while_each_thread(current, tsk);
1647 read_unlock(&tasklist_lock);
1650 retval = wo->notask_error;
1651 if (!retval && !(wo->wo_flags & WNOHANG)) {
1652 retval = -ERESTARTSYS;
1653 if (!signal_pending(current)) {
1659 __set_current_state(TASK_RUNNING);
1660 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1664 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1665 infop, int, options, struct rusage __user *, ru)
1667 struct wait_opts wo;
1668 struct pid *pid = NULL;
1672 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1674 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1687 type = PIDTYPE_PGID;
1695 if (type < PIDTYPE_MAX)
1696 pid = find_get_pid(upid);
1700 wo.wo_flags = options;
1710 * For a WNOHANG return, clear out all the fields
1711 * we would set so the user can easily tell the
1715 ret = put_user(0, &infop->si_signo);
1717 ret = put_user(0, &infop->si_errno);
1719 ret = put_user(0, &infop->si_code);
1721 ret = put_user(0, &infop->si_pid);
1723 ret = put_user(0, &infop->si_uid);
1725 ret = put_user(0, &infop->si_status);
1730 /* avoid REGPARM breakage on x86: */
1731 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1735 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1736 int, options, struct rusage __user *, ru)
1738 struct wait_opts wo;
1739 struct pid *pid = NULL;
1743 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1744 __WNOTHREAD|__WCLONE|__WALL))
1749 else if (upid < 0) {
1750 type = PIDTYPE_PGID;
1751 pid = find_get_pid(-upid);
1752 } else if (upid == 0) {
1753 type = PIDTYPE_PGID;
1754 pid = get_task_pid(current, PIDTYPE_PGID);
1755 } else /* upid > 0 */ {
1757 pid = find_get_pid(upid);
1762 wo.wo_flags = options | WEXITED;
1764 wo.wo_stat = stat_addr;
1769 /* avoid REGPARM breakage on x86: */
1770 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1774 #ifdef __ARCH_WANT_SYS_WAITPID
1777 * sys_waitpid() remains for compatibility. waitpid() should be
1778 * implemented by calling sys_wait4() from libc.a.
1780 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1782 return sys_wait4(pid, stat_addr, options, NULL);