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) {
192 BUG_ON(task_detached(leader));
193 do_notify_parent(leader, leader->exit_signal);
195 * If we were the last child thread and the leader has
196 * exited already, and the leader's parent ignores SIGCHLD,
197 * then we are the one who should release the leader.
199 * do_notify_parent() will have marked it self-reaping in
202 zap_leader = task_detached(leader);
205 * This maintains the invariant that release_task()
206 * only runs on a task in EXIT_DEAD, just for sanity.
209 leader->exit_state = EXIT_DEAD;
212 write_unlock_irq(&tasklist_lock);
214 call_rcu(&p->rcu, delayed_put_task_struct);
217 if (unlikely(zap_leader))
222 * This checks not only the pgrp, but falls back on the pid if no
223 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
226 * The caller must hold rcu lock or the tasklist lock.
228 struct pid *session_of_pgrp(struct pid *pgrp)
230 struct task_struct *p;
231 struct pid *sid = NULL;
233 p = pid_task(pgrp, PIDTYPE_PGID);
235 p = pid_task(pgrp, PIDTYPE_PID);
237 sid = task_session(p);
243 * Determine if a process group is "orphaned", according to the POSIX
244 * definition in 2.2.2.52. Orphaned process groups are not to be affected
245 * by terminal-generated stop signals. Newly orphaned process groups are
246 * to receive a SIGHUP and a SIGCONT.
248 * "I ask you, have you ever known what it is to be an orphan?"
250 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
252 struct task_struct *p;
254 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
255 if ((p == ignored_task) ||
256 (p->exit_state && thread_group_empty(p)) ||
257 is_global_init(p->real_parent))
260 if (task_pgrp(p->real_parent) != pgrp &&
261 task_session(p->real_parent) == task_session(p))
263 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
268 int is_current_pgrp_orphaned(void)
272 read_lock(&tasklist_lock);
273 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
274 read_unlock(&tasklist_lock);
279 static int has_stopped_jobs(struct pid *pgrp)
282 struct task_struct *p;
284 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
285 if (!task_is_stopped(p))
289 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
294 * Check to see if any process groups have become orphaned as
295 * a result of our exiting, and if they have any stopped jobs,
296 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
299 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
301 struct pid *pgrp = task_pgrp(tsk);
302 struct task_struct *ignored_task = tsk;
305 /* exit: our father is in a different pgrp than
306 * we are and we were the only connection outside.
308 parent = tsk->real_parent;
310 /* reparent: our child is in a different pgrp than
311 * we are, and it was the only connection outside.
315 if (task_pgrp(parent) != pgrp &&
316 task_session(parent) == task_session(tsk) &&
317 will_become_orphaned_pgrp(pgrp, ignored_task) &&
318 has_stopped_jobs(pgrp)) {
319 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
320 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
325 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
327 * If a kernel thread is launched as a result of a system call, or if
328 * it ever exits, it should generally reparent itself to kthreadd so it
329 * isn't in the way of other processes and is correctly cleaned up on exit.
331 * The various task state such as scheduling policy and priority may have
332 * been inherited from a user process, so we reset them to sane values here.
334 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
336 static void reparent_to_kthreadd(void)
338 write_lock_irq(&tasklist_lock);
340 ptrace_unlink(current);
341 /* Reparent to init */
342 current->real_parent = current->parent = kthreadd_task;
343 list_move_tail(¤t->sibling, ¤t->real_parent->children);
345 /* Set the exit signal to SIGCHLD so we signal init on exit */
346 current->exit_signal = SIGCHLD;
348 if (task_nice(current) < 0)
349 set_user_nice(current, 0);
353 memcpy(current->signal->rlim, init_task.signal->rlim,
354 sizeof(current->signal->rlim));
356 atomic_inc(&init_cred.usage);
357 commit_creds(&init_cred);
358 write_unlock_irq(&tasklist_lock);
361 void __set_special_pids(struct pid *pid)
363 struct task_struct *curr = current->group_leader;
365 if (task_session(curr) != pid)
366 change_pid(curr, PIDTYPE_SID, pid);
368 if (task_pgrp(curr) != pid)
369 change_pid(curr, PIDTYPE_PGID, pid);
372 static void set_special_pids(struct pid *pid)
374 write_lock_irq(&tasklist_lock);
375 __set_special_pids(pid);
376 write_unlock_irq(&tasklist_lock);
380 * Let kernel threads use this to say that they allow a certain signal.
381 * Must not be used if kthread was cloned with CLONE_SIGHAND.
383 int allow_signal(int sig)
385 if (!valid_signal(sig) || sig < 1)
388 spin_lock_irq(¤t->sighand->siglock);
389 /* This is only needed for daemonize()'ed kthreads */
390 sigdelset(¤t->blocked, sig);
392 * Kernel threads handle their own signals. Let the signal code
393 * know it'll be handled, so that they don't get converted to
394 * SIGKILL or just silently dropped.
396 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
398 spin_unlock_irq(¤t->sighand->siglock);
402 EXPORT_SYMBOL(allow_signal);
404 int disallow_signal(int sig)
406 if (!valid_signal(sig) || sig < 1)
409 spin_lock_irq(¤t->sighand->siglock);
410 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
412 spin_unlock_irq(¤t->sighand->siglock);
416 EXPORT_SYMBOL(disallow_signal);
419 * Put all the gunge required to become a kernel thread without
420 * attached user resources in one place where it belongs.
423 void daemonize(const char *name, ...)
428 va_start(args, name);
429 vsnprintf(current->comm, sizeof(current->comm), name, args);
433 * If we were started as result of loading a module, close all of the
434 * user space pages. We don't need them, and if we didn't close them
435 * they would be locked into memory.
439 * We don't want to have TIF_FREEZE set if the system-wide hibernation
440 * or suspend transition begins right now.
442 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
444 if (current->nsproxy != &init_nsproxy) {
445 get_nsproxy(&init_nsproxy);
446 switch_task_namespaces(current, &init_nsproxy);
448 set_special_pids(&init_struct_pid);
449 proc_clear_tty(current);
451 /* Block and flush all signals */
452 sigfillset(&blocked);
453 sigprocmask(SIG_BLOCK, &blocked, NULL);
454 flush_signals(current);
456 /* Become as one with the init task */
458 daemonize_fs_struct();
460 current->files = init_task.files;
461 atomic_inc(¤t->files->count);
463 reparent_to_kthreadd();
466 EXPORT_SYMBOL(daemonize);
468 static void close_files(struct files_struct * files)
476 * It is safe to dereference the fd table without RCU or
477 * ->file_lock because this is the last reference to the
478 * files structure. But use RCU to shut RCU-lockdep up.
481 fdt = files_fdtable(files);
486 if (i >= fdt->max_fds)
488 set = fdt->open_fds->fds_bits[j++];
491 struct file * file = xchg(&fdt->fd[i], NULL);
493 filp_close(file, files);
503 struct files_struct *get_files_struct(struct task_struct *task)
505 struct files_struct *files;
510 atomic_inc(&files->count);
516 void put_files_struct(struct files_struct *files)
520 if (atomic_dec_and_test(&files->count)) {
523 * Free the fd and fdset arrays if we expanded them.
524 * If the fdtable was embedded, pass files for freeing
525 * at the end of the RCU grace period. Otherwise,
526 * you can free files immediately.
529 fdt = files_fdtable(files);
530 if (fdt != &files->fdtab)
531 kmem_cache_free(files_cachep, files);
537 void reset_files_struct(struct files_struct *files)
539 struct task_struct *tsk = current;
540 struct files_struct *old;
546 put_files_struct(old);
549 void exit_files(struct task_struct *tsk)
551 struct files_struct * files = tsk->files;
557 put_files_struct(files);
561 #ifdef CONFIG_MM_OWNER
563 * Task p is exiting and it owned mm, lets find a new owner for it
566 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
569 * If there are other users of the mm and the owner (us) is exiting
570 * we need to find a new owner to take on the responsibility.
572 if (atomic_read(&mm->mm_users) <= 1)
579 void mm_update_next_owner(struct mm_struct *mm)
581 struct task_struct *c, *g, *p = current;
584 if (!mm_need_new_owner(mm, p))
587 read_lock(&tasklist_lock);
589 * Search in the children
591 list_for_each_entry(c, &p->children, sibling) {
593 goto assign_new_owner;
597 * Search in the siblings
599 list_for_each_entry(c, &p->real_parent->children, sibling) {
601 goto assign_new_owner;
605 * Search through everything else. We should not get
608 do_each_thread(g, c) {
610 goto assign_new_owner;
611 } while_each_thread(g, c);
613 read_unlock(&tasklist_lock);
615 * We found no owner yet mm_users > 1: this implies that we are
616 * most likely racing with swapoff (try_to_unuse()) or /proc or
617 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
626 * The task_lock protects c->mm from changing.
627 * We always want mm->owner->mm == mm
631 * Delay read_unlock() till we have the task_lock()
632 * to ensure that c does not slip away underneath us
634 read_unlock(&tasklist_lock);
644 #endif /* CONFIG_MM_OWNER */
647 * Turn us into a lazy TLB process if we
650 static void exit_mm(struct task_struct * tsk)
652 struct mm_struct *mm = tsk->mm;
653 struct core_state *core_state;
659 * Serialize with any possible pending coredump.
660 * We must hold mmap_sem around checking core_state
661 * and clearing tsk->mm. The core-inducing thread
662 * will increment ->nr_threads for each thread in the
663 * group with ->mm != NULL.
665 down_read(&mm->mmap_sem);
666 core_state = mm->core_state;
668 struct core_thread self;
669 up_read(&mm->mmap_sem);
672 self.next = xchg(&core_state->dumper.next, &self);
674 * Implies mb(), the result of xchg() must be visible
675 * to core_state->dumper.
677 if (atomic_dec_and_test(&core_state->nr_threads))
678 complete(&core_state->startup);
681 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
682 if (!self.task) /* see coredump_finish() */
686 __set_task_state(tsk, TASK_RUNNING);
687 down_read(&mm->mmap_sem);
689 atomic_inc(&mm->mm_count);
690 BUG_ON(mm != tsk->active_mm);
691 /* more a memory barrier than a real lock */
694 up_read(&mm->mmap_sem);
695 enter_lazy_tlb(mm, current);
696 /* We don't want this task to be frozen prematurely */
697 clear_freeze_flag(tsk);
698 if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
699 atomic_dec(&mm->oom_disable_count);
701 mm_update_next_owner(mm);
706 * When we die, we re-parent all our children.
707 * Try to give them to another thread in our thread
708 * group, and if no such member exists, give it to
709 * the child reaper process (ie "init") in our pid
712 static struct task_struct *find_new_reaper(struct task_struct *father)
713 __releases(&tasklist_lock)
714 __acquires(&tasklist_lock)
716 struct pid_namespace *pid_ns = task_active_pid_ns(father);
717 struct task_struct *thread;
720 while_each_thread(father, thread) {
721 if (thread->flags & PF_EXITING)
723 if (unlikely(pid_ns->child_reaper == father))
724 pid_ns->child_reaper = thread;
728 if (unlikely(pid_ns->child_reaper == father)) {
729 write_unlock_irq(&tasklist_lock);
730 if (unlikely(pid_ns == &init_pid_ns))
731 panic("Attempted to kill init!");
733 zap_pid_ns_processes(pid_ns);
734 write_lock_irq(&tasklist_lock);
736 * We can not clear ->child_reaper or leave it alone.
737 * There may by stealth EXIT_DEAD tasks on ->children,
738 * forget_original_parent() must move them somewhere.
740 pid_ns->child_reaper = init_pid_ns.child_reaper;
743 return pid_ns->child_reaper;
747 * Any that need to be release_task'd are put on the @dead list.
749 static void reparent_leader(struct task_struct *father, struct task_struct *p,
750 struct list_head *dead)
752 list_move_tail(&p->sibling, &p->real_parent->children);
754 if (task_detached(p))
757 * If this is a threaded reparent there is no need to
758 * notify anyone anything has happened.
760 if (same_thread_group(p->real_parent, father))
763 /* We don't want people slaying init. */
764 p->exit_signal = SIGCHLD;
766 /* If it has exited notify the new parent about this child's death. */
768 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
769 do_notify_parent(p, p->exit_signal);
770 if (task_detached(p)) {
771 p->exit_state = EXIT_DEAD;
772 list_move_tail(&p->sibling, dead);
776 kill_orphaned_pgrp(p, father);
779 static void forget_original_parent(struct task_struct *father)
781 struct task_struct *p, *n, *reaper;
782 LIST_HEAD(dead_children);
784 write_lock_irq(&tasklist_lock);
786 * Note that exit_ptrace() and find_new_reaper() might
787 * drop tasklist_lock and reacquire it.
790 reaper = find_new_reaper(father);
792 list_for_each_entry_safe(p, n, &father->children, sibling) {
793 struct task_struct *t = p;
795 t->real_parent = reaper;
796 if (t->parent == father) {
798 t->parent = t->real_parent;
800 if (t->pdeath_signal)
801 group_send_sig_info(t->pdeath_signal,
803 } while_each_thread(p, t);
804 reparent_leader(father, p, &dead_children);
806 write_unlock_irq(&tasklist_lock);
808 BUG_ON(!list_empty(&father->children));
810 list_for_each_entry_safe(p, n, &dead_children, sibling) {
811 list_del_init(&p->sibling);
817 * Send signals to all our closest relatives so that they know
818 * to properly mourn us..
820 static void exit_notify(struct task_struct *tsk, int group_dead)
827 * This does two things:
829 * A. Make init inherit all the child processes
830 * B. Check to see if any process groups have become orphaned
831 * as a result of our exiting, and if they have any stopped
832 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
834 forget_original_parent(tsk);
835 exit_task_namespaces(tsk);
837 write_lock_irq(&tasklist_lock);
839 kill_orphaned_pgrp(tsk->group_leader, NULL);
841 /* Let father know we died
843 * Thread signals are configurable, but you aren't going to use
844 * that to send signals to arbitrary processes.
845 * That stops right now.
847 * If the parent exec id doesn't match the exec id we saved
848 * when we started then we know the parent has changed security
851 * If our self_exec id doesn't match our parent_exec_id then
852 * we have changed execution domain as these two values started
853 * the same after a fork.
855 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
856 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
857 tsk->self_exec_id != tsk->parent_exec_id))
858 tsk->exit_signal = SIGCHLD;
860 signal = tracehook_notify_death(tsk, &cookie, group_dead);
862 autoreap = do_notify_parent(tsk, signal);
864 autoreap = (signal == DEATH_REAP);
866 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
868 /* mt-exec, de_thread() is waiting for group leader */
869 if (unlikely(tsk->signal->notify_count < 0))
870 wake_up_process(tsk->signal->group_exit_task);
871 write_unlock_irq(&tasklist_lock);
873 /* If the process is dead, release it - nobody will wait for it */
878 #ifdef CONFIG_DEBUG_STACK_USAGE
879 static void check_stack_usage(void)
881 static DEFINE_SPINLOCK(low_water_lock);
882 static int lowest_to_date = THREAD_SIZE;
885 free = stack_not_used(current);
887 if (free >= lowest_to_date)
890 spin_lock(&low_water_lock);
891 if (free < lowest_to_date) {
892 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
894 current->comm, free);
895 lowest_to_date = free;
897 spin_unlock(&low_water_lock);
900 static inline void check_stack_usage(void) {}
903 NORET_TYPE void do_exit(long code)
905 struct task_struct *tsk = current;
908 profile_task_exit(tsk);
910 WARN_ON(atomic_read(&tsk->fs_excl));
911 WARN_ON(blk_needs_flush_plug(tsk));
913 if (unlikely(in_interrupt()))
914 panic("Aiee, killing interrupt handler!");
915 if (unlikely(!tsk->pid))
916 panic("Attempted to kill the idle task!");
919 * If do_exit is called because this processes oopsed, it's possible
920 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
921 * continuing. Amongst other possible reasons, this is to prevent
922 * mm_release()->clear_child_tid() from writing to a user-controlled
927 ptrace_event(PTRACE_EVENT_EXIT, code);
929 validate_creds_for_do_exit(tsk);
932 * We're taking recursive faults here in do_exit. Safest is to just
933 * leave this task alone and wait for reboot.
935 if (unlikely(tsk->flags & PF_EXITING)) {
937 "Fixing recursive fault but reboot is needed!\n");
939 * We can do this unlocked here. The futex code uses
940 * this flag just to verify whether the pi state
941 * cleanup has been done or not. In the worst case it
942 * loops once more. We pretend that the cleanup was
943 * done as there is no way to return. Either the
944 * OWNER_DIED bit is set by now or we push the blocked
945 * task into the wait for ever nirwana as well.
947 tsk->flags |= PF_EXITPIDONE;
948 set_current_state(TASK_UNINTERRUPTIBLE);
954 exit_signals(tsk); /* sets PF_EXITING */
956 * tsk->flags are checked in the futex code to protect against
957 * an exiting task cleaning up the robust pi futexes.
960 raw_spin_unlock_wait(&tsk->pi_lock);
962 if (unlikely(in_atomic()))
963 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
964 current->comm, task_pid_nr(current),
967 acct_update_integrals(tsk);
968 /* sync mm's RSS info before statistics gathering */
970 sync_mm_rss(tsk, tsk->mm);
971 group_dead = atomic_dec_and_test(&tsk->signal->live);
973 hrtimer_cancel(&tsk->signal->real_timer);
974 exit_itimers(tsk->signal);
976 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
978 acct_collect(code, group_dead);
981 if (unlikely(tsk->audit_context))
984 tsk->exit_code = code;
985 taskstats_exit(tsk, group_dead);
991 trace_sched_process_exit(tsk);
1000 * Flush inherited counters to the parent - before the parent
1001 * gets woken up by child-exit notifications.
1003 * because of cgroup mode, must be called before cgroup_exit()
1005 perf_event_exit_task(tsk);
1007 cgroup_exit(tsk, 1);
1010 disassociate_ctty(1);
1012 module_put(task_thread_info(tsk)->exec_domain->module);
1014 proc_exit_connector(tsk);
1017 * FIXME: do that only when needed, using sched_exit tracepoint
1019 ptrace_put_breakpoints(tsk);
1021 exit_notify(tsk, group_dead);
1024 mpol_put(tsk->mempolicy);
1025 tsk->mempolicy = NULL;
1029 if (unlikely(current->pi_state_cache))
1030 kfree(current->pi_state_cache);
1033 * Make sure we are holding no locks:
1035 debug_check_no_locks_held(tsk);
1037 * We can do this unlocked here. The futex code uses this flag
1038 * just to verify whether the pi state cleanup has been done
1039 * or not. In the worst case it loops once more.
1041 tsk->flags |= PF_EXITPIDONE;
1043 if (tsk->io_context)
1044 exit_io_context(tsk);
1046 if (tsk->splice_pipe)
1047 __free_pipe_info(tsk->splice_pipe);
1049 validate_creds_for_do_exit(tsk);
1053 /* causes final put_task_struct in finish_task_switch(). */
1054 tsk->state = TASK_DEAD;
1057 /* Avoid "noreturn function does return". */
1059 cpu_relax(); /* For when BUG is null */
1062 EXPORT_SYMBOL_GPL(do_exit);
1064 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1072 EXPORT_SYMBOL(complete_and_exit);
1074 SYSCALL_DEFINE1(exit, int, error_code)
1076 do_exit((error_code&0xff)<<8);
1080 * Take down every thread in the group. This is called by fatal signals
1081 * as well as by sys_exit_group (below).
1084 do_group_exit(int exit_code)
1086 struct signal_struct *sig = current->signal;
1088 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1090 if (signal_group_exit(sig))
1091 exit_code = sig->group_exit_code;
1092 else if (!thread_group_empty(current)) {
1093 struct sighand_struct *const sighand = current->sighand;
1094 spin_lock_irq(&sighand->siglock);
1095 if (signal_group_exit(sig))
1096 /* Another thread got here before we took the lock. */
1097 exit_code = sig->group_exit_code;
1099 sig->group_exit_code = exit_code;
1100 sig->flags = SIGNAL_GROUP_EXIT;
1101 zap_other_threads(current);
1103 spin_unlock_irq(&sighand->siglock);
1111 * this kills every thread in the thread group. Note that any externally
1112 * wait4()-ing process will get the correct exit code - even if this
1113 * thread is not the thread group leader.
1115 SYSCALL_DEFINE1(exit_group, int, error_code)
1117 do_group_exit((error_code & 0xff) << 8);
1123 enum pid_type wo_type;
1127 struct siginfo __user *wo_info;
1128 int __user *wo_stat;
1129 struct rusage __user *wo_rusage;
1131 wait_queue_t child_wait;
1136 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1138 if (type != PIDTYPE_PID)
1139 task = task->group_leader;
1140 return task->pids[type].pid;
1143 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1145 return wo->wo_type == PIDTYPE_MAX ||
1146 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1149 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1151 if (!eligible_pid(wo, p))
1153 /* Wait for all children (clone and not) if __WALL is set;
1154 * otherwise, wait for clone children *only* if __WCLONE is
1155 * set; otherwise, wait for non-clone children *only*. (Note:
1156 * A "clone" child here is one that reports to its parent
1157 * using a signal other than SIGCHLD.) */
1158 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1159 && !(wo->wo_flags & __WALL))
1165 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1166 pid_t pid, uid_t uid, int why, int status)
1168 struct siginfo __user *infop;
1169 int retval = wo->wo_rusage
1170 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1173 infop = wo->wo_info;
1176 retval = put_user(SIGCHLD, &infop->si_signo);
1178 retval = put_user(0, &infop->si_errno);
1180 retval = put_user((short)why, &infop->si_code);
1182 retval = put_user(pid, &infop->si_pid);
1184 retval = put_user(uid, &infop->si_uid);
1186 retval = put_user(status, &infop->si_status);
1194 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1195 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1196 * the lock and this task is uninteresting. If we return nonzero, we have
1197 * released the lock and the system call should return.
1199 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1201 unsigned long state;
1202 int retval, status, traced;
1203 pid_t pid = task_pid_vnr(p);
1204 uid_t uid = __task_cred(p)->uid;
1205 struct siginfo __user *infop;
1207 if (!likely(wo->wo_flags & WEXITED))
1210 if (unlikely(wo->wo_flags & WNOWAIT)) {
1211 int exit_code = p->exit_code;
1215 read_unlock(&tasklist_lock);
1216 if ((exit_code & 0x7f) == 0) {
1218 status = exit_code >> 8;
1220 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1221 status = exit_code & 0x7f;
1223 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1227 * Try to move the task's state to DEAD
1228 * only one thread is allowed to do this:
1230 state = xchg(&p->exit_state, EXIT_DEAD);
1231 if (state != EXIT_ZOMBIE) {
1232 BUG_ON(state != EXIT_DEAD);
1236 traced = ptrace_reparented(p);
1238 * It can be ptraced but not reparented, check
1239 * !task_detached() to filter out sub-threads.
1241 if (likely(!traced) && likely(!task_detached(p))) {
1242 struct signal_struct *psig;
1243 struct signal_struct *sig;
1244 unsigned long maxrss;
1245 cputime_t tgutime, tgstime;
1248 * The resource counters for the group leader are in its
1249 * own task_struct. Those for dead threads in the group
1250 * are in its signal_struct, as are those for the child
1251 * processes it has previously reaped. All these
1252 * accumulate in the parent's signal_struct c* fields.
1254 * We don't bother to take a lock here to protect these
1255 * p->signal fields, because they are only touched by
1256 * __exit_signal, which runs with tasklist_lock
1257 * write-locked anyway, and so is excluded here. We do
1258 * need to protect the access to parent->signal fields,
1259 * as other threads in the parent group can be right
1260 * here reaping other children at the same time.
1262 * We use thread_group_times() to get times for the thread
1263 * group, which consolidates times for all threads in the
1264 * group including the group leader.
1266 thread_group_times(p, &tgutime, &tgstime);
1267 spin_lock_irq(&p->real_parent->sighand->siglock);
1268 psig = p->real_parent->signal;
1271 cputime_add(psig->cutime,
1272 cputime_add(tgutime,
1275 cputime_add(psig->cstime,
1276 cputime_add(tgstime,
1279 cputime_add(psig->cgtime,
1280 cputime_add(p->gtime,
1281 cputime_add(sig->gtime,
1284 p->min_flt + sig->min_flt + sig->cmin_flt;
1286 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1288 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1290 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1292 task_io_get_inblock(p) +
1293 sig->inblock + sig->cinblock;
1295 task_io_get_oublock(p) +
1296 sig->oublock + sig->coublock;
1297 maxrss = max(sig->maxrss, sig->cmaxrss);
1298 if (psig->cmaxrss < maxrss)
1299 psig->cmaxrss = maxrss;
1300 task_io_accounting_add(&psig->ioac, &p->ioac);
1301 task_io_accounting_add(&psig->ioac, &sig->ioac);
1302 spin_unlock_irq(&p->real_parent->sighand->siglock);
1306 * Now we are sure this task is interesting, and no other
1307 * thread can reap it because we set its state to EXIT_DEAD.
1309 read_unlock(&tasklist_lock);
1311 retval = wo->wo_rusage
1312 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1313 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1314 ? p->signal->group_exit_code : p->exit_code;
1315 if (!retval && wo->wo_stat)
1316 retval = put_user(status, wo->wo_stat);
1318 infop = wo->wo_info;
1319 if (!retval && infop)
1320 retval = put_user(SIGCHLD, &infop->si_signo);
1321 if (!retval && infop)
1322 retval = put_user(0, &infop->si_errno);
1323 if (!retval && infop) {
1326 if ((status & 0x7f) == 0) {
1330 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1333 retval = put_user((short)why, &infop->si_code);
1335 retval = put_user(status, &infop->si_status);
1337 if (!retval && infop)
1338 retval = put_user(pid, &infop->si_pid);
1339 if (!retval && infop)
1340 retval = put_user(uid, &infop->si_uid);
1345 write_lock_irq(&tasklist_lock);
1346 /* We dropped tasklist, ptracer could die and untrace */
1349 * If this is not a detached task, notify the parent.
1350 * If it's still not detached after that, don't release
1353 if (!task_detached(p)) {
1354 do_notify_parent(p, p->exit_signal);
1355 if (!task_detached(p)) {
1356 p->exit_state = EXIT_ZOMBIE;
1360 write_unlock_irq(&tasklist_lock);
1368 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1371 if (task_is_stopped_or_traced(p) &&
1372 !(p->jobctl & JOBCTL_LISTENING))
1373 return &p->exit_code;
1375 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1376 return &p->signal->group_exit_code;
1382 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1384 * @ptrace: is the wait for ptrace
1385 * @p: task to wait for
1387 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1390 * read_lock(&tasklist_lock), which is released if return value is
1391 * non-zero. Also, grabs and releases @p->sighand->siglock.
1394 * 0 if wait condition didn't exist and search for other wait conditions
1395 * should continue. Non-zero return, -errno on failure and @p's pid on
1396 * success, implies that tasklist_lock is released and wait condition
1397 * search should terminate.
1399 static int wait_task_stopped(struct wait_opts *wo,
1400 int ptrace, struct task_struct *p)
1402 struct siginfo __user *infop;
1403 int retval, exit_code, *p_code, why;
1404 uid_t uid = 0; /* unneeded, required by compiler */
1408 * Traditionally we see ptrace'd stopped tasks regardless of options.
1410 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1413 if (!task_stopped_code(p, ptrace))
1417 spin_lock_irq(&p->sighand->siglock);
1419 p_code = task_stopped_code(p, ptrace);
1420 if (unlikely(!p_code))
1423 exit_code = *p_code;
1427 if (!unlikely(wo->wo_flags & WNOWAIT))
1432 spin_unlock_irq(&p->sighand->siglock);
1437 * Now we are pretty sure this task is interesting.
1438 * Make sure it doesn't get reaped out from under us while we
1439 * give up the lock and then examine it below. We don't want to
1440 * keep holding onto the tasklist_lock while we call getrusage and
1441 * possibly take page faults for user memory.
1444 pid = task_pid_vnr(p);
1445 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1446 read_unlock(&tasklist_lock);
1448 if (unlikely(wo->wo_flags & WNOWAIT))
1449 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1451 retval = wo->wo_rusage
1452 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1453 if (!retval && wo->wo_stat)
1454 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1456 infop = wo->wo_info;
1457 if (!retval && infop)
1458 retval = put_user(SIGCHLD, &infop->si_signo);
1459 if (!retval && infop)
1460 retval = put_user(0, &infop->si_errno);
1461 if (!retval && infop)
1462 retval = put_user((short)why, &infop->si_code);
1463 if (!retval && infop)
1464 retval = put_user(exit_code, &infop->si_status);
1465 if (!retval && infop)
1466 retval = put_user(pid, &infop->si_pid);
1467 if (!retval && infop)
1468 retval = put_user(uid, &infop->si_uid);
1478 * Handle do_wait work for one task in a live, non-stopped state.
1479 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1480 * the lock and this task is uninteresting. If we return nonzero, we have
1481 * released the lock and the system call should return.
1483 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1489 if (!unlikely(wo->wo_flags & WCONTINUED))
1492 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1495 spin_lock_irq(&p->sighand->siglock);
1496 /* Re-check with the lock held. */
1497 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1498 spin_unlock_irq(&p->sighand->siglock);
1501 if (!unlikely(wo->wo_flags & WNOWAIT))
1502 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1504 spin_unlock_irq(&p->sighand->siglock);
1506 pid = task_pid_vnr(p);
1508 read_unlock(&tasklist_lock);
1511 retval = wo->wo_rusage
1512 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1514 if (!retval && wo->wo_stat)
1515 retval = put_user(0xffff, wo->wo_stat);
1519 retval = wait_noreap_copyout(wo, p, pid, uid,
1520 CLD_CONTINUED, SIGCONT);
1521 BUG_ON(retval == 0);
1528 * Consider @p for a wait by @parent.
1530 * -ECHILD should be in ->notask_error before the first call.
1531 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1532 * Returns zero if the search for a child should continue;
1533 * then ->notask_error is 0 if @p is an eligible child,
1534 * or another error from security_task_wait(), or still -ECHILD.
1536 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1537 struct task_struct *p)
1539 int ret = eligible_child(wo, p);
1543 ret = security_task_wait(p);
1544 if (unlikely(ret < 0)) {
1546 * If we have not yet seen any eligible child,
1547 * then let this error code replace -ECHILD.
1548 * A permission error will give the user a clue
1549 * to look for security policy problems, rather
1550 * than for mysterious wait bugs.
1552 if (wo->notask_error)
1553 wo->notask_error = ret;
1557 /* dead body doesn't have much to contribute */
1558 if (p->exit_state == EXIT_DEAD)
1562 if (p->exit_state == EXIT_ZOMBIE) {
1564 * A zombie ptracee is only visible to its ptracer.
1565 * Notification and reaping will be cascaded to the real
1566 * parent when the ptracer detaches.
1568 if (likely(!ptrace) && unlikely(p->ptrace)) {
1569 /* it will become visible, clear notask_error */
1570 wo->notask_error = 0;
1574 /* we don't reap group leaders with subthreads */
1575 if (!delay_group_leader(p))
1576 return wait_task_zombie(wo, p);
1579 * Allow access to stopped/continued state via zombie by
1580 * falling through. Clearing of notask_error is complex.
1584 * If WEXITED is set, notask_error should naturally be
1585 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1586 * so, if there are live subthreads, there are events to
1587 * wait for. If all subthreads are dead, it's still safe
1588 * to clear - this function will be called again in finite
1589 * amount time once all the subthreads are released and
1590 * will then return without clearing.
1594 * Stopped state is per-task and thus can't change once the
1595 * target task dies. Only continued and exited can happen.
1596 * Clear notask_error if WCONTINUED | WEXITED.
1598 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1599 wo->notask_error = 0;
1602 * If @p is ptraced by a task in its real parent's group,
1603 * hide group stop/continued state when looking at @p as
1604 * the real parent; otherwise, a single stop can be
1605 * reported twice as group and ptrace stops.
1607 * If a ptracer wants to distinguish the two events for its
1608 * own children, it should create a separate process which
1609 * takes the role of real parent.
1611 if (likely(!ptrace) && p->ptrace &&
1612 same_thread_group(p->parent, p->real_parent))
1616 * @p is alive and it's gonna stop, continue or exit, so
1617 * there always is something to wait for.
1619 wo->notask_error = 0;
1623 * Wait for stopped. Depending on @ptrace, different stopped state
1624 * is used and the two don't interact with each other.
1626 ret = wait_task_stopped(wo, ptrace, p);
1631 * Wait for continued. There's only one continued state and the
1632 * ptracer can consume it which can confuse the real parent. Don't
1633 * use WCONTINUED from ptracer. You don't need or want it.
1635 return wait_task_continued(wo, p);
1639 * Do the work of do_wait() for one thread in the group, @tsk.
1641 * -ECHILD should be in ->notask_error before the first call.
1642 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1643 * Returns zero if the search for a child should continue; then
1644 * ->notask_error is 0 if there were any eligible children,
1645 * or another error from security_task_wait(), or still -ECHILD.
1647 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1649 struct task_struct *p;
1651 list_for_each_entry(p, &tsk->children, sibling) {
1652 int ret = wait_consider_task(wo, 0, p);
1660 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1662 struct task_struct *p;
1664 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1665 int ret = wait_consider_task(wo, 1, p);
1673 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1674 int sync, void *key)
1676 struct wait_opts *wo = container_of(wait, struct wait_opts,
1678 struct task_struct *p = key;
1680 if (!eligible_pid(wo, p))
1683 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1686 return default_wake_function(wait, mode, sync, key);
1689 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1691 __wake_up_sync_key(&parent->signal->wait_chldexit,
1692 TASK_INTERRUPTIBLE, 1, p);
1695 static long do_wait(struct wait_opts *wo)
1697 struct task_struct *tsk;
1700 trace_sched_process_wait(wo->wo_pid);
1702 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1703 wo->child_wait.private = current;
1704 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1707 * If there is nothing that can match our critiera just get out.
1708 * We will clear ->notask_error to zero if we see any child that
1709 * might later match our criteria, even if we are not able to reap
1712 wo->notask_error = -ECHILD;
1713 if ((wo->wo_type < PIDTYPE_MAX) &&
1714 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1717 set_current_state(TASK_INTERRUPTIBLE);
1718 read_lock(&tasklist_lock);
1721 retval = do_wait_thread(wo, tsk);
1725 retval = ptrace_do_wait(wo, tsk);
1729 if (wo->wo_flags & __WNOTHREAD)
1731 } while_each_thread(current, tsk);
1732 read_unlock(&tasklist_lock);
1735 retval = wo->notask_error;
1736 if (!retval && !(wo->wo_flags & WNOHANG)) {
1737 retval = -ERESTARTSYS;
1738 if (!signal_pending(current)) {
1744 __set_current_state(TASK_RUNNING);
1745 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1749 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1750 infop, int, options, struct rusage __user *, ru)
1752 struct wait_opts wo;
1753 struct pid *pid = NULL;
1757 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1759 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1772 type = PIDTYPE_PGID;
1780 if (type < PIDTYPE_MAX)
1781 pid = find_get_pid(upid);
1785 wo.wo_flags = options;
1795 * For a WNOHANG return, clear out all the fields
1796 * we would set so the user can easily tell the
1800 ret = put_user(0, &infop->si_signo);
1802 ret = put_user(0, &infop->si_errno);
1804 ret = put_user(0, &infop->si_code);
1806 ret = put_user(0, &infop->si_pid);
1808 ret = put_user(0, &infop->si_uid);
1810 ret = put_user(0, &infop->si_status);
1815 /* avoid REGPARM breakage on x86: */
1816 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1820 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1821 int, options, struct rusage __user *, ru)
1823 struct wait_opts wo;
1824 struct pid *pid = NULL;
1828 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1829 __WNOTHREAD|__WCLONE|__WALL))
1834 else if (upid < 0) {
1835 type = PIDTYPE_PGID;
1836 pid = find_get_pid(-upid);
1837 } else if (upid == 0) {
1838 type = PIDTYPE_PGID;
1839 pid = get_task_pid(current, PIDTYPE_PGID);
1840 } else /* upid > 0 */ {
1842 pid = find_get_pid(upid);
1847 wo.wo_flags = options | WEXITED;
1849 wo.wo_stat = stat_addr;
1854 /* avoid REGPARM breakage on x86: */
1855 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1859 #ifdef __ARCH_WANT_SYS_WAITPID
1862 * sys_waitpid() remains for compatibility. waitpid() should be
1863 * implemented by calling sys_wait4() from libc.a.
1865 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1867 return sys_wait4(pid, stat_addr, options, NULL);