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>
53 #include <linux/oom.h>
55 #include <asm/uaccess.h>
56 #include <asm/unistd.h>
57 #include <asm/pgtable.h>
58 #include <asm/mmu_context.h>
60 static void exit_mm(struct task_struct * tsk);
62 static void __unhash_process(struct task_struct *p, bool group_dead)
65 detach_pid(p, PIDTYPE_PID);
67 detach_pid(p, PIDTYPE_PGID);
68 detach_pid(p, PIDTYPE_SID);
70 list_del_rcu(&p->tasks);
71 list_del_init(&p->sibling);
72 __this_cpu_dec(process_counts);
74 list_del_rcu(&p->thread_group);
78 * This function expects the tasklist_lock write-locked.
80 static void __exit_signal(struct task_struct *tsk)
82 struct signal_struct *sig = tsk->signal;
83 bool group_dead = thread_group_leader(tsk);
84 struct sighand_struct *sighand;
85 struct tty_struct *uninitialized_var(tty);
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);
94 posix_cpu_timers_exit_group(tsk);
99 * This can only happen if the caller is de_thread().
100 * FIXME: this is the temporary hack, we should teach
101 * posix-cpu-timers to handle this case correctly.
103 if (unlikely(has_group_leader_pid(tsk)))
104 posix_cpu_timers_exit_group(tsk);
107 * If there is any task waiting for the group exit
110 if (sig->notify_count > 0 && !--sig->notify_count)
111 wake_up_process(sig->group_exit_task);
113 if (tsk == sig->curr_target)
114 sig->curr_target = next_thread(tsk);
116 * Accumulate here the counters for all threads but the
117 * group leader as they die, so they can be added into
118 * the process-wide totals when those are taken.
119 * The group leader stays around as a zombie as long
120 * as there are other threads. When it gets reaped,
121 * the exit.c code will add its counts into these totals.
122 * We won't ever get here for the group leader, since it
123 * will have been the last reference on the signal_struct.
125 sig->utime = cputime_add(sig->utime, tsk->utime);
126 sig->stime = cputime_add(sig->stime, tsk->stime);
127 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
128 sig->min_flt += tsk->min_flt;
129 sig->maj_flt += tsk->maj_flt;
130 sig->nvcsw += tsk->nvcsw;
131 sig->nivcsw += tsk->nivcsw;
132 sig->inblock += task_io_get_inblock(tsk);
133 sig->oublock += task_io_get_oublock(tsk);
134 task_io_accounting_add(&sig->ioac, &tsk->ioac);
135 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
139 __unhash_process(tsk, group_dead);
142 * Do this under ->siglock, we can race with another thread
143 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
145 flush_sigqueue(&tsk->pending);
147 spin_unlock(&sighand->siglock);
149 __cleanup_sighand(sighand);
150 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
152 flush_sigqueue(&sig->shared_pending);
157 static void delayed_put_task_struct(struct rcu_head *rhp)
159 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161 perf_event_delayed_put(tsk);
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 /* don't need to get the RCU readlock here - the process is dead and
173 * can't be modifying its own credentials. But shut RCU-lockdep up */
175 atomic_dec(&__task_cred(p)->user->processes);
180 write_lock_irq(&tasklist_lock);
181 ptrace_release_task(p);
185 * If we are the last non-leader member of the thread
186 * group, and the leader is zombie, then notify the
187 * group leader's parent process. (if it wants notification.)
190 leader = p->group_leader;
191 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
193 * If we were the last child thread and the leader has
194 * exited already, and the leader's parent ignores SIGCHLD,
195 * then we are the one who should release the leader.
197 zap_leader = do_notify_parent(leader, leader->exit_signal);
199 leader->exit_state = EXIT_DEAD;
202 write_unlock_irq(&tasklist_lock);
204 call_rcu(&p->rcu, delayed_put_task_struct);
207 if (unlikely(zap_leader))
212 * This checks not only the pgrp, but falls back on the pid if no
213 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
216 * The caller must hold rcu lock or the tasklist lock.
218 struct pid *session_of_pgrp(struct pid *pgrp)
220 struct task_struct *p;
221 struct pid *sid = NULL;
223 p = pid_task(pgrp, PIDTYPE_PGID);
225 p = pid_task(pgrp, PIDTYPE_PID);
227 sid = task_session(p);
233 * Determine if a process group is "orphaned", according to the POSIX
234 * definition in 2.2.2.52. Orphaned process groups are not to be affected
235 * by terminal-generated stop signals. Newly orphaned process groups are
236 * to receive a SIGHUP and a SIGCONT.
238 * "I ask you, have you ever known what it is to be an orphan?"
240 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
242 struct task_struct *p;
244 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
245 if ((p == ignored_task) ||
246 (p->exit_state && thread_group_empty(p)) ||
247 is_global_init(p->real_parent))
250 if (task_pgrp(p->real_parent) != pgrp &&
251 task_session(p->real_parent) == task_session(p))
253 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
258 int is_current_pgrp_orphaned(void)
262 read_lock(&tasklist_lock);
263 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
264 read_unlock(&tasklist_lock);
269 static int has_stopped_jobs(struct pid *pgrp)
272 struct task_struct *p;
274 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
275 if (!task_is_stopped(p))
279 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
284 * Check to see if any process groups have become orphaned as
285 * a result of our exiting, and if they have any stopped jobs,
286 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
289 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
291 struct pid *pgrp = task_pgrp(tsk);
292 struct task_struct *ignored_task = tsk;
295 /* exit: our father is in a different pgrp than
296 * we are and we were the only connection outside.
298 parent = tsk->real_parent;
300 /* reparent: our child is in a different pgrp than
301 * we are, and it was the only connection outside.
305 if (task_pgrp(parent) != pgrp &&
306 task_session(parent) == task_session(tsk) &&
307 will_become_orphaned_pgrp(pgrp, ignored_task) &&
308 has_stopped_jobs(pgrp)) {
309 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
310 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
315 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
317 * If a kernel thread is launched as a result of a system call, or if
318 * it ever exits, it should generally reparent itself to kthreadd so it
319 * isn't in the way of other processes and is correctly cleaned up on exit.
321 * The various task state such as scheduling policy and priority may have
322 * been inherited from a user process, so we reset them to sane values here.
324 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
326 static void reparent_to_kthreadd(void)
328 write_lock_irq(&tasklist_lock);
330 ptrace_unlink(current);
331 /* Reparent to init */
332 current->real_parent = current->parent = kthreadd_task;
333 list_move_tail(¤t->sibling, ¤t->real_parent->children);
335 /* Set the exit signal to SIGCHLD so we signal init on exit */
336 current->exit_signal = SIGCHLD;
338 if (task_nice(current) < 0)
339 set_user_nice(current, 0);
343 memcpy(current->signal->rlim, init_task.signal->rlim,
344 sizeof(current->signal->rlim));
346 atomic_inc(&init_cred.usage);
347 commit_creds(&init_cred);
348 write_unlock_irq(&tasklist_lock);
351 void __set_special_pids(struct pid *pid)
353 struct task_struct *curr = current->group_leader;
355 if (task_session(curr) != pid)
356 change_pid(curr, PIDTYPE_SID, pid);
358 if (task_pgrp(curr) != pid)
359 change_pid(curr, PIDTYPE_PGID, pid);
362 static void set_special_pids(struct pid *pid)
364 write_lock_irq(&tasklist_lock);
365 __set_special_pids(pid);
366 write_unlock_irq(&tasklist_lock);
370 * Let kernel threads use this to say that they allow a certain signal.
371 * Must not be used if kthread was cloned with CLONE_SIGHAND.
373 int allow_signal(int sig)
375 if (!valid_signal(sig) || sig < 1)
378 spin_lock_irq(¤t->sighand->siglock);
379 /* This is only needed for daemonize()'ed kthreads */
380 sigdelset(¤t->blocked, sig);
382 * Kernel threads handle their own signals. Let the signal code
383 * know it'll be handled, so that they don't get converted to
384 * SIGKILL or just silently dropped.
386 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
388 spin_unlock_irq(¤t->sighand->siglock);
392 EXPORT_SYMBOL(allow_signal);
394 int disallow_signal(int sig)
396 if (!valid_signal(sig) || sig < 1)
399 spin_lock_irq(¤t->sighand->siglock);
400 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
402 spin_unlock_irq(¤t->sighand->siglock);
406 EXPORT_SYMBOL(disallow_signal);
409 * Put all the gunge required to become a kernel thread without
410 * attached user resources in one place where it belongs.
413 void daemonize(const char *name, ...)
418 va_start(args, name);
419 vsnprintf(current->comm, sizeof(current->comm), name, args);
423 * If we were started as result of loading a module, close all of the
424 * user space pages. We don't need them, and if we didn't close them
425 * they would be locked into memory.
429 * We don't want to have TIF_FREEZE set if the system-wide hibernation
430 * or suspend transition begins right now.
432 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
434 if (current->nsproxy != &init_nsproxy) {
435 get_nsproxy(&init_nsproxy);
436 switch_task_namespaces(current, &init_nsproxy);
438 set_special_pids(&init_struct_pid);
439 proc_clear_tty(current);
441 /* Block and flush all signals */
442 sigfillset(&blocked);
443 sigprocmask(SIG_BLOCK, &blocked, NULL);
444 flush_signals(current);
446 /* Become as one with the init task */
448 daemonize_fs_struct();
450 current->files = init_task.files;
451 atomic_inc(¤t->files->count);
453 reparent_to_kthreadd();
456 EXPORT_SYMBOL(daemonize);
458 static void close_files(struct files_struct * files)
466 * It is safe to dereference the fd table without RCU or
467 * ->file_lock because this is the last reference to the
468 * files structure. But use RCU to shut RCU-lockdep up.
471 fdt = files_fdtable(files);
476 if (i >= fdt->max_fds)
478 set = fdt->open_fds->fds_bits[j++];
481 struct file * file = xchg(&fdt->fd[i], NULL);
483 filp_close(file, files);
493 struct files_struct *get_files_struct(struct task_struct *task)
495 struct files_struct *files;
500 atomic_inc(&files->count);
506 void put_files_struct(struct files_struct *files)
510 if (atomic_dec_and_test(&files->count)) {
513 * Free the fd and fdset arrays if we expanded them.
514 * If the fdtable was embedded, pass files for freeing
515 * at the end of the RCU grace period. Otherwise,
516 * you can free files immediately.
519 fdt = files_fdtable(files);
520 if (fdt != &files->fdtab)
521 kmem_cache_free(files_cachep, files);
527 void reset_files_struct(struct files_struct *files)
529 struct task_struct *tsk = current;
530 struct files_struct *old;
536 put_files_struct(old);
539 void exit_files(struct task_struct *tsk)
541 struct files_struct * files = tsk->files;
547 put_files_struct(files);
551 #ifdef CONFIG_MM_OWNER
553 * Task p is exiting and it owned mm, lets find a new owner for it
556 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
559 * If there are other users of the mm and the owner (us) is exiting
560 * we need to find a new owner to take on the responsibility.
562 if (atomic_read(&mm->mm_users) <= 1)
569 void mm_update_next_owner(struct mm_struct *mm)
571 struct task_struct *c, *g, *p = current;
574 if (!mm_need_new_owner(mm, p))
577 read_lock(&tasklist_lock);
579 * Search in the children
581 list_for_each_entry(c, &p->children, sibling) {
583 goto assign_new_owner;
587 * Search in the siblings
589 list_for_each_entry(c, &p->real_parent->children, sibling) {
591 goto assign_new_owner;
595 * Search through everything else. We should not get
598 do_each_thread(g, c) {
600 goto assign_new_owner;
601 } while_each_thread(g, c);
603 read_unlock(&tasklist_lock);
605 * We found no owner yet mm_users > 1: this implies that we are
606 * most likely racing with swapoff (try_to_unuse()) or /proc or
607 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
616 * The task_lock protects c->mm from changing.
617 * We always want mm->owner->mm == mm
621 * Delay read_unlock() till we have the task_lock()
622 * to ensure that c does not slip away underneath us
624 read_unlock(&tasklist_lock);
634 #endif /* CONFIG_MM_OWNER */
637 * Turn us into a lazy TLB process if we
640 static void exit_mm(struct task_struct * tsk)
642 struct mm_struct *mm = tsk->mm;
643 struct core_state *core_state;
649 * Serialize with any possible pending coredump.
650 * We must hold mmap_sem around checking core_state
651 * and clearing tsk->mm. The core-inducing thread
652 * will increment ->nr_threads for each thread in the
653 * group with ->mm != NULL.
655 down_read(&mm->mmap_sem);
656 core_state = mm->core_state;
658 struct core_thread self;
659 up_read(&mm->mmap_sem);
662 self.next = xchg(&core_state->dumper.next, &self);
664 * Implies mb(), the result of xchg() must be visible
665 * to core_state->dumper.
667 if (atomic_dec_and_test(&core_state->nr_threads))
668 complete(&core_state->startup);
671 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
672 if (!self.task) /* see coredump_finish() */
676 __set_task_state(tsk, TASK_RUNNING);
677 down_read(&mm->mmap_sem);
679 atomic_inc(&mm->mm_count);
680 BUG_ON(mm != tsk->active_mm);
681 /* more a memory barrier than a real lock */
684 up_read(&mm->mmap_sem);
685 enter_lazy_tlb(mm, current);
686 /* We don't want this task to be frozen prematurely */
687 clear_freeze_flag(tsk);
688 if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
689 atomic_dec(&mm->oom_disable_count);
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)
703 __releases(&tasklist_lock)
704 __acquires(&tasklist_lock)
706 struct pid_namespace *pid_ns = task_active_pid_ns(father);
707 struct task_struct *thread;
710 while_each_thread(father, thread) {
711 if (thread->flags & PF_EXITING)
713 if (unlikely(pid_ns->child_reaper == father))
714 pid_ns->child_reaper = thread;
718 if (unlikely(pid_ns->child_reaper == father)) {
719 write_unlock_irq(&tasklist_lock);
720 if (unlikely(pid_ns == &init_pid_ns))
721 panic("Attempted to kill init!");
723 zap_pid_ns_processes(pid_ns);
724 write_lock_irq(&tasklist_lock);
726 * We can not clear ->child_reaper or leave it alone.
727 * There may by stealth EXIT_DEAD tasks on ->children,
728 * forget_original_parent() must move them somewhere.
730 pid_ns->child_reaper = init_pid_ns.child_reaper;
733 return pid_ns->child_reaper;
737 * Any that need to be release_task'd are put on the @dead list.
739 static void reparent_leader(struct task_struct *father, struct task_struct *p,
740 struct list_head *dead)
742 list_move_tail(&p->sibling, &p->real_parent->children);
744 if (p->exit_state == EXIT_DEAD)
747 * If this is a threaded reparent there is no need to
748 * notify anyone anything has happened.
750 if (same_thread_group(p->real_parent, father))
753 /* We don't want people slaying init. */
754 p->exit_signal = SIGCHLD;
756 /* If it has exited notify the new parent about this child's death. */
758 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
759 if (do_notify_parent(p, p->exit_signal)) {
760 p->exit_state = EXIT_DEAD;
761 list_move_tail(&p->sibling, dead);
765 kill_orphaned_pgrp(p, father);
768 static void forget_original_parent(struct task_struct *father)
770 struct task_struct *p, *n, *reaper;
771 LIST_HEAD(dead_children);
773 write_lock_irq(&tasklist_lock);
775 * Note that exit_ptrace() and find_new_reaper() might
776 * drop tasklist_lock and reacquire it.
779 reaper = find_new_reaper(father);
781 list_for_each_entry_safe(p, n, &father->children, sibling) {
782 struct task_struct *t = p;
784 t->real_parent = reaper;
785 if (t->parent == father) {
787 t->parent = t->real_parent;
789 if (t->pdeath_signal)
790 group_send_sig_info(t->pdeath_signal,
792 } while_each_thread(p, t);
793 reparent_leader(father, p, &dead_children);
795 write_unlock_irq(&tasklist_lock);
797 BUG_ON(!list_empty(&father->children));
799 list_for_each_entry_safe(p, n, &dead_children, sibling) {
800 list_del_init(&p->sibling);
806 * Send signals to all our closest relatives so that they know
807 * to properly mourn us..
809 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 arbitrary 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 (thread_group_leader(tsk) && tsk->exit_signal != SIGCHLD &&
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 if (unlikely(tsk->ptrace)) {
848 int sig = thread_group_leader(tsk) &&
849 thread_group_empty(tsk) &&
850 !ptrace_reparented(tsk) ?
851 tsk->exit_signal : SIGCHLD;
852 autoreap = do_notify_parent(tsk, sig);
853 } else if (thread_group_leader(tsk)) {
854 autoreap = thread_group_empty(tsk) &&
855 do_notify_parent(tsk, tsk->exit_signal);
860 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
862 /* mt-exec, de_thread() is waiting for group leader */
863 if (unlikely(tsk->signal->notify_count < 0))
864 wake_up_process(tsk->signal->group_exit_task);
865 write_unlock_irq(&tasklist_lock);
867 /* If the process is dead, release it - nobody will wait for it */
872 #ifdef CONFIG_DEBUG_STACK_USAGE
873 static void check_stack_usage(void)
875 static DEFINE_SPINLOCK(low_water_lock);
876 static int lowest_to_date = THREAD_SIZE;
879 free = stack_not_used(current);
881 if (free >= lowest_to_date)
884 spin_lock(&low_water_lock);
885 if (free < lowest_to_date) {
886 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
888 current->comm, free);
889 lowest_to_date = free;
891 spin_unlock(&low_water_lock);
894 static inline void check_stack_usage(void) {}
897 NORET_TYPE void do_exit(long code)
899 struct task_struct *tsk = current;
902 profile_task_exit(tsk);
904 WARN_ON(atomic_read(&tsk->fs_excl));
905 WARN_ON(blk_needs_flush_plug(tsk));
907 if (unlikely(in_interrupt()))
908 panic("Aiee, killing interrupt handler!");
909 if (unlikely(!tsk->pid))
910 panic("Attempted to kill the idle task!");
913 * If do_exit is called because this processes oopsed, it's possible
914 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
915 * continuing. Amongst other possible reasons, this is to prevent
916 * mm_release()->clear_child_tid() from writing to a user-controlled
921 ptrace_event(PTRACE_EVENT_EXIT, code);
923 validate_creds_for_do_exit(tsk);
926 * We're taking recursive faults here in do_exit. Safest is to just
927 * leave this task alone and wait for reboot.
929 if (unlikely(tsk->flags & PF_EXITING)) {
931 "Fixing recursive fault but reboot is needed!\n");
933 * We can do this unlocked here. The futex code uses
934 * this flag just to verify whether the pi state
935 * cleanup has been done or not. In the worst case it
936 * loops once more. We pretend that the cleanup was
937 * done as there is no way to return. Either the
938 * OWNER_DIED bit is set by now or we push the blocked
939 * task into the wait for ever nirwana as well.
941 tsk->flags |= PF_EXITPIDONE;
942 set_current_state(TASK_UNINTERRUPTIBLE);
948 exit_signals(tsk); /* sets PF_EXITING */
950 * tsk->flags are checked in the futex code to protect against
951 * an exiting task cleaning up the robust pi futexes.
954 raw_spin_unlock_wait(&tsk->pi_lock);
956 if (unlikely(in_atomic()))
957 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
958 current->comm, task_pid_nr(current),
961 acct_update_integrals(tsk);
962 /* sync mm's RSS info before statistics gathering */
964 sync_mm_rss(tsk, tsk->mm);
965 group_dead = atomic_dec_and_test(&tsk->signal->live);
967 hrtimer_cancel(&tsk->signal->real_timer);
968 exit_itimers(tsk->signal);
970 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
972 acct_collect(code, group_dead);
975 if (unlikely(tsk->audit_context))
978 tsk->exit_code = code;
979 taskstats_exit(tsk, group_dead);
985 trace_sched_process_exit(tsk);
994 * Flush inherited counters to the parent - before the parent
995 * gets woken up by child-exit notifications.
997 * because of cgroup mode, must be called before cgroup_exit()
999 perf_event_exit_task(tsk);
1001 cgroup_exit(tsk, 1);
1004 disassociate_ctty(1);
1006 module_put(task_thread_info(tsk)->exec_domain->module);
1008 proc_exit_connector(tsk);
1011 * FIXME: do that only when needed, using sched_exit tracepoint
1013 ptrace_put_breakpoints(tsk);
1015 exit_notify(tsk, group_dead);
1018 mpol_put(tsk->mempolicy);
1019 tsk->mempolicy = NULL;
1023 if (unlikely(current->pi_state_cache))
1024 kfree(current->pi_state_cache);
1027 * Make sure we are holding no locks:
1029 debug_check_no_locks_held(tsk);
1031 * We can do this unlocked here. The futex code uses this flag
1032 * just to verify whether the pi state cleanup has been done
1033 * or not. In the worst case it loops once more.
1035 tsk->flags |= PF_EXITPIDONE;
1037 if (tsk->io_context)
1038 exit_io_context(tsk);
1040 if (tsk->splice_pipe)
1041 __free_pipe_info(tsk->splice_pipe);
1043 validate_creds_for_do_exit(tsk);
1047 /* causes final put_task_struct in finish_task_switch(). */
1048 tsk->state = TASK_DEAD;
1051 /* Avoid "noreturn function does return". */
1053 cpu_relax(); /* For when BUG is null */
1056 EXPORT_SYMBOL_GPL(do_exit);
1058 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1066 EXPORT_SYMBOL(complete_and_exit);
1068 SYSCALL_DEFINE1(exit, int, error_code)
1070 do_exit((error_code&0xff)<<8);
1074 * Take down every thread in the group. This is called by fatal signals
1075 * as well as by sys_exit_group (below).
1078 do_group_exit(int exit_code)
1080 struct signal_struct *sig = current->signal;
1082 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1084 if (signal_group_exit(sig))
1085 exit_code = sig->group_exit_code;
1086 else if (!thread_group_empty(current)) {
1087 struct sighand_struct *const sighand = current->sighand;
1088 spin_lock_irq(&sighand->siglock);
1089 if (signal_group_exit(sig))
1090 /* Another thread got here before we took the lock. */
1091 exit_code = sig->group_exit_code;
1093 sig->group_exit_code = exit_code;
1094 sig->flags = SIGNAL_GROUP_EXIT;
1095 zap_other_threads(current);
1097 spin_unlock_irq(&sighand->siglock);
1105 * this kills every thread in the thread group. Note that any externally
1106 * wait4()-ing process will get the correct exit code - even if this
1107 * thread is not the thread group leader.
1109 SYSCALL_DEFINE1(exit_group, int, error_code)
1111 do_group_exit((error_code & 0xff) << 8);
1117 enum pid_type wo_type;
1121 struct siginfo __user *wo_info;
1122 int __user *wo_stat;
1123 struct rusage __user *wo_rusage;
1125 wait_queue_t child_wait;
1130 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1132 if (type != PIDTYPE_PID)
1133 task = task->group_leader;
1134 return task->pids[type].pid;
1137 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1139 return wo->wo_type == PIDTYPE_MAX ||
1140 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1143 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1145 if (!eligible_pid(wo, p))
1147 /* Wait for all children (clone and not) if __WALL is set;
1148 * otherwise, wait for clone children *only* if __WCLONE is
1149 * set; otherwise, wait for non-clone children *only*. (Note:
1150 * A "clone" child here is one that reports to its parent
1151 * using a signal other than SIGCHLD.) */
1152 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1153 && !(wo->wo_flags & __WALL))
1159 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1160 pid_t pid, uid_t uid, int why, int status)
1162 struct siginfo __user *infop;
1163 int retval = wo->wo_rusage
1164 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1167 infop = wo->wo_info;
1170 retval = put_user(SIGCHLD, &infop->si_signo);
1172 retval = put_user(0, &infop->si_errno);
1174 retval = put_user((short)why, &infop->si_code);
1176 retval = put_user(pid, &infop->si_pid);
1178 retval = put_user(uid, &infop->si_uid);
1180 retval = put_user(status, &infop->si_status);
1188 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1189 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1190 * the lock and this task is uninteresting. If we return nonzero, we have
1191 * released the lock and the system call should return.
1193 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1195 unsigned long state;
1196 int retval, status, traced;
1197 pid_t pid = task_pid_vnr(p);
1198 uid_t uid = __task_cred(p)->uid;
1199 struct siginfo __user *infop;
1201 if (!likely(wo->wo_flags & WEXITED))
1204 if (unlikely(wo->wo_flags & WNOWAIT)) {
1205 int exit_code = p->exit_code;
1209 read_unlock(&tasklist_lock);
1210 if ((exit_code & 0x7f) == 0) {
1212 status = exit_code >> 8;
1214 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1215 status = exit_code & 0x7f;
1217 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1221 * Try to move the task's state to DEAD
1222 * only one thread is allowed to do this:
1224 state = xchg(&p->exit_state, EXIT_DEAD);
1225 if (state != EXIT_ZOMBIE) {
1226 BUG_ON(state != EXIT_DEAD);
1230 traced = ptrace_reparented(p);
1232 * It can be ptraced but not reparented, check
1233 * thread_group_leader() to filter out sub-threads.
1235 if (likely(!traced) && thread_group_leader(p)) {
1236 struct signal_struct *psig;
1237 struct signal_struct *sig;
1238 unsigned long maxrss;
1239 cputime_t tgutime, tgstime;
1242 * The resource counters for the group leader are in its
1243 * own task_struct. Those for dead threads in the group
1244 * are in its signal_struct, as are those for the child
1245 * processes it has previously reaped. All these
1246 * accumulate in the parent's signal_struct c* fields.
1248 * We don't bother to take a lock here to protect these
1249 * p->signal fields, because they are only touched by
1250 * __exit_signal, which runs with tasklist_lock
1251 * write-locked anyway, and so is excluded here. We do
1252 * need to protect the access to parent->signal fields,
1253 * as other threads in the parent group can be right
1254 * here reaping other children at the same time.
1256 * We use thread_group_times() to get times for the thread
1257 * group, which consolidates times for all threads in the
1258 * group including the group leader.
1260 thread_group_times(p, &tgutime, &tgstime);
1261 spin_lock_irq(&p->real_parent->sighand->siglock);
1262 psig = p->real_parent->signal;
1265 cputime_add(psig->cutime,
1266 cputime_add(tgutime,
1269 cputime_add(psig->cstime,
1270 cputime_add(tgstime,
1273 cputime_add(psig->cgtime,
1274 cputime_add(p->gtime,
1275 cputime_add(sig->gtime,
1278 p->min_flt + sig->min_flt + sig->cmin_flt;
1280 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1282 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1284 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1286 task_io_get_inblock(p) +
1287 sig->inblock + sig->cinblock;
1289 task_io_get_oublock(p) +
1290 sig->oublock + sig->coublock;
1291 maxrss = max(sig->maxrss, sig->cmaxrss);
1292 if (psig->cmaxrss < maxrss)
1293 psig->cmaxrss = maxrss;
1294 task_io_accounting_add(&psig->ioac, &p->ioac);
1295 task_io_accounting_add(&psig->ioac, &sig->ioac);
1296 spin_unlock_irq(&p->real_parent->sighand->siglock);
1300 * Now we are sure this task is interesting, and no other
1301 * thread can reap it because we set its state to EXIT_DEAD.
1303 read_unlock(&tasklist_lock);
1305 retval = wo->wo_rusage
1306 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1307 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1308 ? p->signal->group_exit_code : p->exit_code;
1309 if (!retval && wo->wo_stat)
1310 retval = put_user(status, wo->wo_stat);
1312 infop = wo->wo_info;
1313 if (!retval && infop)
1314 retval = put_user(SIGCHLD, &infop->si_signo);
1315 if (!retval && infop)
1316 retval = put_user(0, &infop->si_errno);
1317 if (!retval && infop) {
1320 if ((status & 0x7f) == 0) {
1324 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1327 retval = put_user((short)why, &infop->si_code);
1329 retval = put_user(status, &infop->si_status);
1331 if (!retval && infop)
1332 retval = put_user(pid, &infop->si_pid);
1333 if (!retval && infop)
1334 retval = put_user(uid, &infop->si_uid);
1339 write_lock_irq(&tasklist_lock);
1340 /* We dropped tasklist, ptracer could die and untrace */
1343 * If this is not a sub-thread, notify the parent.
1344 * If parent wants a zombie, don't release it now.
1346 if (thread_group_leader(p) &&
1347 !do_notify_parent(p, p->exit_signal)) {
1348 p->exit_state = EXIT_ZOMBIE;
1351 write_unlock_irq(&tasklist_lock);
1359 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1362 if (task_is_stopped_or_traced(p) &&
1363 !(p->jobctl & JOBCTL_LISTENING))
1364 return &p->exit_code;
1366 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1367 return &p->signal->group_exit_code;
1373 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1375 * @ptrace: is the wait for ptrace
1376 * @p: task to wait for
1378 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1381 * read_lock(&tasklist_lock), which is released if return value is
1382 * non-zero. Also, grabs and releases @p->sighand->siglock.
1385 * 0 if wait condition didn't exist and search for other wait conditions
1386 * should continue. Non-zero return, -errno on failure and @p's pid on
1387 * success, implies that tasklist_lock is released and wait condition
1388 * search should terminate.
1390 static int wait_task_stopped(struct wait_opts *wo,
1391 int ptrace, struct task_struct *p)
1393 struct siginfo __user *infop;
1394 int retval, exit_code, *p_code, why;
1395 uid_t uid = 0; /* unneeded, required by compiler */
1399 * Traditionally we see ptrace'd stopped tasks regardless of options.
1401 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1404 if (!task_stopped_code(p, ptrace))
1408 spin_lock_irq(&p->sighand->siglock);
1410 p_code = task_stopped_code(p, ptrace);
1411 if (unlikely(!p_code))
1414 exit_code = *p_code;
1418 if (!unlikely(wo->wo_flags & WNOWAIT))
1423 spin_unlock_irq(&p->sighand->siglock);
1428 * Now we are pretty sure this task is interesting.
1429 * Make sure it doesn't get reaped out from under us while we
1430 * give up the lock and then examine it below. We don't want to
1431 * keep holding onto the tasklist_lock while we call getrusage and
1432 * possibly take page faults for user memory.
1435 pid = task_pid_vnr(p);
1436 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1437 read_unlock(&tasklist_lock);
1439 if (unlikely(wo->wo_flags & WNOWAIT))
1440 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1442 retval = wo->wo_rusage
1443 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1444 if (!retval && wo->wo_stat)
1445 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1447 infop = wo->wo_info;
1448 if (!retval && infop)
1449 retval = put_user(SIGCHLD, &infop->si_signo);
1450 if (!retval && infop)
1451 retval = put_user(0, &infop->si_errno);
1452 if (!retval && infop)
1453 retval = put_user((short)why, &infop->si_code);
1454 if (!retval && infop)
1455 retval = put_user(exit_code, &infop->si_status);
1456 if (!retval && infop)
1457 retval = put_user(pid, &infop->si_pid);
1458 if (!retval && infop)
1459 retval = put_user(uid, &infop->si_uid);
1469 * Handle do_wait work for one task in a live, non-stopped state.
1470 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1471 * the lock and this task is uninteresting. If we return nonzero, we have
1472 * released the lock and the system call should return.
1474 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1480 if (!unlikely(wo->wo_flags & WCONTINUED))
1483 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1486 spin_lock_irq(&p->sighand->siglock);
1487 /* Re-check with the lock held. */
1488 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1489 spin_unlock_irq(&p->sighand->siglock);
1492 if (!unlikely(wo->wo_flags & WNOWAIT))
1493 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1495 spin_unlock_irq(&p->sighand->siglock);
1497 pid = task_pid_vnr(p);
1499 read_unlock(&tasklist_lock);
1502 retval = wo->wo_rusage
1503 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1505 if (!retval && wo->wo_stat)
1506 retval = put_user(0xffff, wo->wo_stat);
1510 retval = wait_noreap_copyout(wo, p, pid, uid,
1511 CLD_CONTINUED, SIGCONT);
1512 BUG_ON(retval == 0);
1519 * Consider @p for a wait by @parent.
1521 * -ECHILD should be in ->notask_error before the first call.
1522 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1523 * Returns zero if the search for a child should continue;
1524 * then ->notask_error is 0 if @p is an eligible child,
1525 * or another error from security_task_wait(), or still -ECHILD.
1527 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1528 struct task_struct *p)
1530 int ret = eligible_child(wo, p);
1534 ret = security_task_wait(p);
1535 if (unlikely(ret < 0)) {
1537 * If we have not yet seen any eligible child,
1538 * then let this error code replace -ECHILD.
1539 * A permission error will give the user a clue
1540 * to look for security policy problems, rather
1541 * than for mysterious wait bugs.
1543 if (wo->notask_error)
1544 wo->notask_error = ret;
1548 /* dead body doesn't have much to contribute */
1549 if (p->exit_state == EXIT_DEAD)
1553 if (p->exit_state == EXIT_ZOMBIE) {
1555 * A zombie ptracee is only visible to its ptracer.
1556 * Notification and reaping will be cascaded to the real
1557 * parent when the ptracer detaches.
1559 if (likely(!ptrace) && unlikely(p->ptrace)) {
1560 /* it will become visible, clear notask_error */
1561 wo->notask_error = 0;
1565 /* we don't reap group leaders with subthreads */
1566 if (!delay_group_leader(p))
1567 return wait_task_zombie(wo, p);
1570 * Allow access to stopped/continued state via zombie by
1571 * falling through. Clearing of notask_error is complex.
1575 * If WEXITED is set, notask_error should naturally be
1576 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1577 * so, if there are live subthreads, there are events to
1578 * wait for. If all subthreads are dead, it's still safe
1579 * to clear - this function will be called again in finite
1580 * amount time once all the subthreads are released and
1581 * will then return without clearing.
1585 * Stopped state is per-task and thus can't change once the
1586 * target task dies. Only continued and exited can happen.
1587 * Clear notask_error if WCONTINUED | WEXITED.
1589 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1590 wo->notask_error = 0;
1593 * If @p is ptraced by a task in its real parent's group,
1594 * hide group stop/continued state when looking at @p as
1595 * the real parent; otherwise, a single stop can be
1596 * reported twice as group and ptrace stops.
1598 * If a ptracer wants to distinguish the two events for its
1599 * own children, it should create a separate process which
1600 * takes the role of real parent.
1602 if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1606 * @p is alive and it's gonna stop, continue or exit, so
1607 * there always is something to wait for.
1609 wo->notask_error = 0;
1613 * Wait for stopped. Depending on @ptrace, different stopped state
1614 * is used and the two don't interact with each other.
1616 ret = wait_task_stopped(wo, ptrace, p);
1621 * Wait for continued. There's only one continued state and the
1622 * ptracer can consume it which can confuse the real parent. Don't
1623 * use WCONTINUED from ptracer. You don't need or want it.
1625 return wait_task_continued(wo, p);
1629 * Do the work of do_wait() for one thread in the group, @tsk.
1631 * -ECHILD should be in ->notask_error before the first call.
1632 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1633 * Returns zero if the search for a child should continue; then
1634 * ->notask_error is 0 if there were any eligible children,
1635 * or another error from security_task_wait(), or still -ECHILD.
1637 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1639 struct task_struct *p;
1641 list_for_each_entry(p, &tsk->children, sibling) {
1642 int ret = wait_consider_task(wo, 0, p);
1650 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1652 struct task_struct *p;
1654 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1655 int ret = wait_consider_task(wo, 1, p);
1663 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1664 int sync, void *key)
1666 struct wait_opts *wo = container_of(wait, struct wait_opts,
1668 struct task_struct *p = key;
1670 if (!eligible_pid(wo, p))
1673 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1676 return default_wake_function(wait, mode, sync, key);
1679 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1681 __wake_up_sync_key(&parent->signal->wait_chldexit,
1682 TASK_INTERRUPTIBLE, 1, p);
1685 static long do_wait(struct wait_opts *wo)
1687 struct task_struct *tsk;
1690 trace_sched_process_wait(wo->wo_pid);
1692 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1693 wo->child_wait.private = current;
1694 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1697 * If there is nothing that can match our critiera just get out.
1698 * We will clear ->notask_error to zero if we see any child that
1699 * might later match our criteria, even if we are not able to reap
1702 wo->notask_error = -ECHILD;
1703 if ((wo->wo_type < PIDTYPE_MAX) &&
1704 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1707 set_current_state(TASK_INTERRUPTIBLE);
1708 read_lock(&tasklist_lock);
1711 retval = do_wait_thread(wo, tsk);
1715 retval = ptrace_do_wait(wo, tsk);
1719 if (wo->wo_flags & __WNOTHREAD)
1721 } while_each_thread(current, tsk);
1722 read_unlock(&tasklist_lock);
1725 retval = wo->notask_error;
1726 if (!retval && !(wo->wo_flags & WNOHANG)) {
1727 retval = -ERESTARTSYS;
1728 if (!signal_pending(current)) {
1734 __set_current_state(TASK_RUNNING);
1735 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1739 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1740 infop, int, options, struct rusage __user *, ru)
1742 struct wait_opts wo;
1743 struct pid *pid = NULL;
1747 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1749 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1762 type = PIDTYPE_PGID;
1770 if (type < PIDTYPE_MAX)
1771 pid = find_get_pid(upid);
1775 wo.wo_flags = options;
1785 * For a WNOHANG return, clear out all the fields
1786 * we would set so the user can easily tell the
1790 ret = put_user(0, &infop->si_signo);
1792 ret = put_user(0, &infop->si_errno);
1794 ret = put_user(0, &infop->si_code);
1796 ret = put_user(0, &infop->si_pid);
1798 ret = put_user(0, &infop->si_uid);
1800 ret = put_user(0, &infop->si_status);
1805 /* avoid REGPARM breakage on x86: */
1806 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1810 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1811 int, options, struct rusage __user *, ru)
1813 struct wait_opts wo;
1814 struct pid *pid = NULL;
1818 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1819 __WNOTHREAD|__WCLONE|__WALL))
1824 else if (upid < 0) {
1825 type = PIDTYPE_PGID;
1826 pid = find_get_pid(-upid);
1827 } else if (upid == 0) {
1828 type = PIDTYPE_PGID;
1829 pid = get_task_pid(current, PIDTYPE_PGID);
1830 } else /* upid > 0 */ {
1832 pid = find_get_pid(upid);
1837 wo.wo_flags = options | WEXITED;
1839 wo.wo_stat = stat_addr;
1844 /* avoid REGPARM breakage on x86: */
1845 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1849 #ifdef __ARCH_WANT_SYS_WAITPID
1852 * sys_waitpid() remains for compatibility. waitpid() should be
1853 * implemented by calling sys_wait4() from libc.a.
1855 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1857 return sys_wait4(pid, stat_addr, options, NULL);