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 lockdep_tasklist_lock_is_held());
89 spin_lock(&sighand->siglock);
91 posix_cpu_timers_exit(tsk);
93 posix_cpu_timers_exit_group(tsk);
98 * This can only happen if the caller is de_thread().
99 * FIXME: this is the temporary hack, we should teach
100 * posix-cpu-timers to handle this case correctly.
102 if (unlikely(has_group_leader_pid(tsk)))
103 posix_cpu_timers_exit_group(tsk);
106 * If there is any task waiting for the group exit
109 if (sig->notify_count > 0 && !--sig->notify_count)
110 wake_up_process(sig->group_exit_task);
112 if (tsk == sig->curr_target)
113 sig->curr_target = next_thread(tsk);
115 * Accumulate here the counters for all threads but the
116 * group leader as they die, so they can be added into
117 * the process-wide totals when those are taken.
118 * The group leader stays around as a zombie as long
119 * as there are other threads. When it gets reaped,
120 * the exit.c code will add its counts into these totals.
121 * We won't ever get here for the group leader, since it
122 * will have been the last reference on the signal_struct.
124 sig->utime = cputime_add(sig->utime, tsk->utime);
125 sig->stime = cputime_add(sig->stime, tsk->stime);
126 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
127 sig->min_flt += tsk->min_flt;
128 sig->maj_flt += tsk->maj_flt;
129 sig->nvcsw += tsk->nvcsw;
130 sig->nivcsw += tsk->nivcsw;
131 sig->inblock += task_io_get_inblock(tsk);
132 sig->oublock += task_io_get_oublock(tsk);
133 task_io_accounting_add(&sig->ioac, &tsk->ioac);
134 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
138 __unhash_process(tsk, group_dead);
141 * Do this under ->siglock, we can race with another thread
142 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
144 flush_sigqueue(&tsk->pending);
146 spin_unlock(&sighand->siglock);
148 __cleanup_sighand(sighand);
149 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
151 flush_sigqueue(&sig->shared_pending);
156 static void delayed_put_task_struct(struct rcu_head *rhp)
158 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
160 perf_event_delayed_put(tsk);
161 trace_sched_process_free(tsk);
162 put_task_struct(tsk);
166 void release_task(struct task_struct * p)
168 struct task_struct *leader;
171 /* don't need to get the RCU readlock here - the process is dead and
172 * can't be modifying its own credentials. But shut RCU-lockdep up */
174 atomic_dec(&__task_cred(p)->user->processes);
179 write_lock_irq(&tasklist_lock);
180 ptrace_release_task(p);
184 * If we are the last non-leader member of the thread
185 * group, and the leader is zombie, then notify the
186 * group leader's parent process. (if it wants notification.)
189 leader = p->group_leader;
190 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
192 * If we were the last child thread and the leader has
193 * exited already, and the leader's parent ignores SIGCHLD,
194 * then we are the one who should release the leader.
196 zap_leader = do_notify_parent(leader, leader->exit_signal);
198 leader->exit_state = EXIT_DEAD;
201 write_unlock_irq(&tasklist_lock);
203 call_rcu(&p->rcu, delayed_put_task_struct);
206 if (unlikely(zap_leader))
211 * This checks not only the pgrp, but falls back on the pid if no
212 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
215 * The caller must hold rcu lock or the tasklist lock.
217 struct pid *session_of_pgrp(struct pid *pgrp)
219 struct task_struct *p;
220 struct pid *sid = NULL;
222 p = pid_task(pgrp, PIDTYPE_PGID);
224 p = pid_task(pgrp, PIDTYPE_PID);
226 sid = task_session(p);
232 * Determine if a process group is "orphaned", according to the POSIX
233 * definition in 2.2.2.52. Orphaned process groups are not to be affected
234 * by terminal-generated stop signals. Newly orphaned process groups are
235 * to receive a SIGHUP and a SIGCONT.
237 * "I ask you, have you ever known what it is to be an orphan?"
239 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
241 struct task_struct *p;
243 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
244 if ((p == ignored_task) ||
245 (p->exit_state && thread_group_empty(p)) ||
246 is_global_init(p->real_parent))
249 if (task_pgrp(p->real_parent) != pgrp &&
250 task_session(p->real_parent) == task_session(p))
252 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
257 int is_current_pgrp_orphaned(void)
261 read_lock(&tasklist_lock);
262 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
263 read_unlock(&tasklist_lock);
268 static bool has_stopped_jobs(struct pid *pgrp)
270 struct task_struct *p;
272 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
273 if (p->signal->flags & SIGNAL_STOP_STOPPED)
275 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
281 * Check to see if any process groups have become orphaned as
282 * a result of our exiting, and if they have any stopped jobs,
283 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
286 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
288 struct pid *pgrp = task_pgrp(tsk);
289 struct task_struct *ignored_task = tsk;
292 /* exit: our father is in a different pgrp than
293 * we are and we were the only connection outside.
295 parent = tsk->real_parent;
297 /* reparent: our child is in a different pgrp than
298 * we are, and it was the only connection outside.
302 if (task_pgrp(parent) != pgrp &&
303 task_session(parent) == task_session(tsk) &&
304 will_become_orphaned_pgrp(pgrp, ignored_task) &&
305 has_stopped_jobs(pgrp)) {
306 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
307 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
312 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
314 * If a kernel thread is launched as a result of a system call, or if
315 * it ever exits, it should generally reparent itself to kthreadd so it
316 * isn't in the way of other processes and is correctly cleaned up on exit.
318 * The various task state such as scheduling policy and priority may have
319 * been inherited from a user process, so we reset them to sane values here.
321 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
323 static void reparent_to_kthreadd(void)
325 write_lock_irq(&tasklist_lock);
327 ptrace_unlink(current);
328 /* Reparent to init */
329 current->real_parent = current->parent = kthreadd_task;
330 list_move_tail(¤t->sibling, ¤t->real_parent->children);
332 /* Set the exit signal to SIGCHLD so we signal init on exit */
333 current->exit_signal = SIGCHLD;
335 if (task_nice(current) < 0)
336 set_user_nice(current, 0);
340 memcpy(current->signal->rlim, init_task.signal->rlim,
341 sizeof(current->signal->rlim));
343 atomic_inc(&init_cred.usage);
344 commit_creds(&init_cred);
345 write_unlock_irq(&tasklist_lock);
348 void __set_special_pids(struct pid *pid)
350 struct task_struct *curr = current->group_leader;
352 if (task_session(curr) != pid)
353 change_pid(curr, PIDTYPE_SID, pid);
355 if (task_pgrp(curr) != pid)
356 change_pid(curr, PIDTYPE_PGID, pid);
359 static void set_special_pids(struct pid *pid)
361 write_lock_irq(&tasklist_lock);
362 __set_special_pids(pid);
363 write_unlock_irq(&tasklist_lock);
367 * Let kernel threads use this to say that they allow a certain signal.
368 * Must not be used if kthread was cloned with CLONE_SIGHAND.
370 int allow_signal(int sig)
372 if (!valid_signal(sig) || sig < 1)
375 spin_lock_irq(¤t->sighand->siglock);
376 /* This is only needed for daemonize()'ed kthreads */
377 sigdelset(¤t->blocked, sig);
379 * Kernel threads handle their own signals. Let the signal code
380 * know it'll be handled, so that they don't get converted to
381 * SIGKILL or just silently dropped.
383 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
385 spin_unlock_irq(¤t->sighand->siglock);
389 EXPORT_SYMBOL(allow_signal);
391 int disallow_signal(int sig)
393 if (!valid_signal(sig) || sig < 1)
396 spin_lock_irq(¤t->sighand->siglock);
397 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
399 spin_unlock_irq(¤t->sighand->siglock);
403 EXPORT_SYMBOL(disallow_signal);
406 * Put all the gunge required to become a kernel thread without
407 * attached user resources in one place where it belongs.
410 void daemonize(const char *name, ...)
415 va_start(args, name);
416 vsnprintf(current->comm, sizeof(current->comm), name, args);
420 * If we were started as result of loading a module, close all of the
421 * user space pages. We don't need them, and if we didn't close them
422 * they would be locked into memory.
426 * We don't want to have TIF_FREEZE set if the system-wide hibernation
427 * or suspend transition begins right now.
429 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
431 if (current->nsproxy != &init_nsproxy) {
432 get_nsproxy(&init_nsproxy);
433 switch_task_namespaces(current, &init_nsproxy);
435 set_special_pids(&init_struct_pid);
436 proc_clear_tty(current);
438 /* Block and flush all signals */
439 sigfillset(&blocked);
440 sigprocmask(SIG_BLOCK, &blocked, NULL);
441 flush_signals(current);
443 /* Become as one with the init task */
445 daemonize_fs_struct();
447 current->files = init_task.files;
448 atomic_inc(¤t->files->count);
450 reparent_to_kthreadd();
453 EXPORT_SYMBOL(daemonize);
455 static void close_files(struct files_struct * files)
463 * It is safe to dereference the fd table without RCU or
464 * ->file_lock because this is the last reference to the
465 * files structure. But use RCU to shut RCU-lockdep up.
468 fdt = files_fdtable(files);
473 if (i >= fdt->max_fds)
475 set = fdt->open_fds->fds_bits[j++];
478 struct file * file = xchg(&fdt->fd[i], NULL);
480 filp_close(file, files);
490 struct files_struct *get_files_struct(struct task_struct *task)
492 struct files_struct *files;
497 atomic_inc(&files->count);
503 void put_files_struct(struct files_struct *files)
507 if (atomic_dec_and_test(&files->count)) {
510 * Free the fd and fdset arrays if we expanded them.
511 * If the fdtable was embedded, pass files for freeing
512 * at the end of the RCU grace period. Otherwise,
513 * you can free files immediately.
516 fdt = files_fdtable(files);
517 if (fdt != &files->fdtab)
518 kmem_cache_free(files_cachep, files);
524 void reset_files_struct(struct files_struct *files)
526 struct task_struct *tsk = current;
527 struct files_struct *old;
533 put_files_struct(old);
536 void exit_files(struct task_struct *tsk)
538 struct files_struct * files = tsk->files;
544 put_files_struct(files);
548 #ifdef CONFIG_MM_OWNER
550 * A task is exiting. If it owned this mm, find a new owner for the mm.
552 void mm_update_next_owner(struct mm_struct *mm)
554 struct task_struct *c, *g, *p = current;
558 * If the exiting or execing task is not the owner, it's
559 * someone else's problem.
564 * The current owner is exiting/execing and there are no other
565 * candidates. Do not leave the mm pointing to a possibly
566 * freed task structure.
568 if (atomic_read(&mm->mm_users) <= 1) {
573 read_lock(&tasklist_lock);
575 * Search in the children
577 list_for_each_entry(c, &p->children, sibling) {
579 goto assign_new_owner;
583 * Search in the siblings
585 list_for_each_entry(c, &p->real_parent->children, sibling) {
587 goto assign_new_owner;
591 * Search through everything else. We should not get
594 do_each_thread(g, c) {
596 goto assign_new_owner;
597 } while_each_thread(g, c);
599 read_unlock(&tasklist_lock);
601 * We found no owner yet mm_users > 1: this implies that we are
602 * most likely racing with swapoff (try_to_unuse()) or /proc or
603 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
612 * The task_lock protects c->mm from changing.
613 * We always want mm->owner->mm == mm
617 * Delay read_unlock() till we have the task_lock()
618 * to ensure that c does not slip away underneath us
620 read_unlock(&tasklist_lock);
630 #endif /* CONFIG_MM_OWNER */
633 * Turn us into a lazy TLB process if we
636 static void exit_mm(struct task_struct * tsk)
638 struct mm_struct *mm = tsk->mm;
639 struct core_state *core_state;
645 * Serialize with any possible pending coredump.
646 * We must hold mmap_sem around checking core_state
647 * and clearing tsk->mm. The core-inducing thread
648 * will increment ->nr_threads for each thread in the
649 * group with ->mm != NULL.
651 down_read(&mm->mmap_sem);
652 core_state = mm->core_state;
654 struct core_thread self;
655 up_read(&mm->mmap_sem);
658 self.next = xchg(&core_state->dumper.next, &self);
660 * Implies mb(), the result of xchg() must be visible
661 * to core_state->dumper.
663 if (atomic_dec_and_test(&core_state->nr_threads))
664 complete(&core_state->startup);
667 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
668 if (!self.task) /* see coredump_finish() */
672 __set_task_state(tsk, TASK_RUNNING);
673 down_read(&mm->mmap_sem);
675 atomic_inc(&mm->mm_count);
676 BUG_ON(mm != tsk->active_mm);
677 /* more a memory barrier than a real lock */
680 up_read(&mm->mmap_sem);
681 enter_lazy_tlb(mm, current);
682 /* We don't want this task to be frozen prematurely */
683 clear_freeze_flag(tsk);
685 mm_update_next_owner(mm);
690 * When we die, we re-parent all our children.
691 * Try to give them to another thread in our thread
692 * group, and if no such member exists, give it to
693 * the child reaper process (ie "init") in our pid
696 static struct task_struct *find_new_reaper(struct task_struct *father)
697 __releases(&tasklist_lock)
698 __acquires(&tasklist_lock)
700 struct pid_namespace *pid_ns = task_active_pid_ns(father);
701 struct task_struct *thread;
704 while_each_thread(father, thread) {
705 if (thread->flags & PF_EXITING)
707 if (unlikely(pid_ns->child_reaper == father))
708 pid_ns->child_reaper = thread;
712 if (unlikely(pid_ns->child_reaper == father)) {
713 write_unlock_irq(&tasklist_lock);
714 if (unlikely(pid_ns == &init_pid_ns))
715 panic("Attempted to kill init!");
717 zap_pid_ns_processes(pid_ns);
718 write_lock_irq(&tasklist_lock);
720 * We can not clear ->child_reaper or leave it alone.
721 * There may by stealth EXIT_DEAD tasks on ->children,
722 * forget_original_parent() must move them somewhere.
724 pid_ns->child_reaper = init_pid_ns.child_reaper;
727 return pid_ns->child_reaper;
731 * Any that need to be release_task'd are put on the @dead list.
733 static void reparent_leader(struct task_struct *father, struct task_struct *p,
734 struct list_head *dead)
736 list_move_tail(&p->sibling, &p->real_parent->children);
738 if (p->exit_state == EXIT_DEAD)
741 * If this is a threaded reparent there is no need to
742 * notify anyone anything has happened.
744 if (same_thread_group(p->real_parent, father))
747 /* We don't want people slaying init. */
748 p->exit_signal = SIGCHLD;
750 /* If it has exited notify the new parent about this child's death. */
752 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
753 if (do_notify_parent(p, p->exit_signal)) {
754 p->exit_state = EXIT_DEAD;
755 list_move_tail(&p->sibling, dead);
759 kill_orphaned_pgrp(p, father);
762 static void forget_original_parent(struct task_struct *father)
764 struct task_struct *p, *n, *reaper;
765 LIST_HEAD(dead_children);
767 write_lock_irq(&tasklist_lock);
769 * Note that exit_ptrace() and find_new_reaper() might
770 * drop tasklist_lock and reacquire it.
773 reaper = find_new_reaper(father);
775 list_for_each_entry_safe(p, n, &father->children, sibling) {
776 struct task_struct *t = p;
778 t->real_parent = reaper;
779 if (t->parent == father) {
781 t->parent = t->real_parent;
783 if (t->pdeath_signal)
784 group_send_sig_info(t->pdeath_signal,
786 } while_each_thread(p, t);
787 reparent_leader(father, p, &dead_children);
789 write_unlock_irq(&tasklist_lock);
791 BUG_ON(!list_empty(&father->children));
793 list_for_each_entry_safe(p, n, &dead_children, sibling) {
794 list_del_init(&p->sibling);
800 * Send signals to all our closest relatives so that they know
801 * to properly mourn us..
803 static void exit_notify(struct task_struct *tsk, int group_dead)
808 * This does two things:
810 * A. Make init inherit all the child processes
811 * B. Check to see if any process groups have become orphaned
812 * as a result of our exiting, and if they have any stopped
813 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
815 forget_original_parent(tsk);
816 exit_task_namespaces(tsk);
818 write_lock_irq(&tasklist_lock);
820 kill_orphaned_pgrp(tsk->group_leader, NULL);
822 /* Let father know we died
824 * Thread signals are configurable, but you aren't going to use
825 * that to send signals to arbitrary processes.
826 * That stops right now.
828 * If the parent exec id doesn't match the exec id we saved
829 * when we started then we know the parent has changed security
832 if (thread_group_leader(tsk) && tsk->exit_signal != SIGCHLD &&
833 tsk->parent_exec_id != tsk->real_parent->self_exec_id)
834 tsk->exit_signal = SIGCHLD;
836 if (unlikely(tsk->ptrace)) {
837 int sig = thread_group_leader(tsk) &&
838 thread_group_empty(tsk) &&
839 !ptrace_reparented(tsk) ?
840 tsk->exit_signal : SIGCHLD;
841 autoreap = do_notify_parent(tsk, sig);
842 } else if (thread_group_leader(tsk)) {
843 autoreap = thread_group_empty(tsk) &&
844 do_notify_parent(tsk, tsk->exit_signal);
849 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
851 /* mt-exec, de_thread() is waiting for group leader */
852 if (unlikely(tsk->signal->notify_count < 0))
853 wake_up_process(tsk->signal->group_exit_task);
854 write_unlock_irq(&tasklist_lock);
856 /* If the process is dead, release it - nobody will wait for it */
861 #ifdef CONFIG_DEBUG_STACK_USAGE
862 static void check_stack_usage(void)
864 static DEFINE_SPINLOCK(low_water_lock);
865 static int lowest_to_date = THREAD_SIZE;
868 free = stack_not_used(current);
870 if (free >= lowest_to_date)
873 spin_lock(&low_water_lock);
874 if (free < lowest_to_date) {
875 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
877 current->comm, free);
878 lowest_to_date = free;
880 spin_unlock(&low_water_lock);
883 static inline void check_stack_usage(void) {}
886 NORET_TYPE void do_exit(long code)
888 struct task_struct *tsk = current;
891 profile_task_exit(tsk);
893 WARN_ON(blk_needs_flush_plug(tsk));
895 if (unlikely(in_interrupt()))
896 panic("Aiee, killing interrupt handler!");
897 if (unlikely(!tsk->pid))
898 panic("Attempted to kill the idle task!");
901 * If do_exit is called because this processes oopsed, it's possible
902 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
903 * continuing. Amongst other possible reasons, this is to prevent
904 * mm_release()->clear_child_tid() from writing to a user-controlled
909 ptrace_event(PTRACE_EVENT_EXIT, code);
911 validate_creds_for_do_exit(tsk);
914 * We're taking recursive faults here in do_exit. Safest is to just
915 * leave this task alone and wait for reboot.
917 if (unlikely(tsk->flags & PF_EXITING)) {
919 "Fixing recursive fault but reboot is needed!\n");
921 * We can do this unlocked here. The futex code uses
922 * this flag just to verify whether the pi state
923 * cleanup has been done or not. In the worst case it
924 * loops once more. We pretend that the cleanup was
925 * done as there is no way to return. Either the
926 * OWNER_DIED bit is set by now or we push the blocked
927 * task into the wait for ever nirwana as well.
929 tsk->flags |= PF_EXITPIDONE;
930 set_current_state(TASK_UNINTERRUPTIBLE);
936 exit_signals(tsk); /* sets PF_EXITING */
938 * tsk->flags are checked in the futex code to protect against
939 * an exiting task cleaning up the robust pi futexes.
942 raw_spin_unlock_wait(&tsk->pi_lock);
944 if (unlikely(in_atomic()))
945 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
946 current->comm, task_pid_nr(current),
949 acct_update_integrals(tsk);
950 /* sync mm's RSS info before statistics gathering */
952 sync_mm_rss(tsk, tsk->mm);
953 group_dead = atomic_dec_and_test(&tsk->signal->live);
955 hrtimer_cancel(&tsk->signal->real_timer);
956 exit_itimers(tsk->signal);
958 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
960 acct_collect(code, group_dead);
963 if (unlikely(tsk->audit_context))
966 tsk->exit_code = code;
967 taskstats_exit(tsk, group_dead);
973 trace_sched_process_exit(tsk);
983 * Flush inherited counters to the parent - before the parent
984 * gets woken up by child-exit notifications.
986 * because of cgroup mode, must be called before cgroup_exit()
988 perf_event_exit_task(tsk);
993 disassociate_ctty(1);
995 module_put(task_thread_info(tsk)->exec_domain->module);
997 proc_exit_connector(tsk);
1000 * FIXME: do that only when needed, using sched_exit tracepoint
1002 ptrace_put_breakpoints(tsk);
1004 exit_notify(tsk, group_dead);
1007 mpol_put(tsk->mempolicy);
1008 tsk->mempolicy = NULL;
1012 if (unlikely(current->pi_state_cache))
1013 kfree(current->pi_state_cache);
1016 * Make sure we are holding no locks:
1018 debug_check_no_locks_held(tsk);
1020 * We can do this unlocked here. The futex code uses this flag
1021 * just to verify whether the pi state cleanup has been done
1022 * or not. In the worst case it loops once more.
1024 tsk->flags |= PF_EXITPIDONE;
1026 if (tsk->io_context)
1027 exit_io_context(tsk);
1029 if (tsk->splice_pipe)
1030 __free_pipe_info(tsk->splice_pipe);
1032 validate_creds_for_do_exit(tsk);
1036 /* causes final put_task_struct in finish_task_switch(). */
1037 tsk->state = TASK_DEAD;
1040 /* Avoid "noreturn function does return". */
1042 cpu_relax(); /* For when BUG is null */
1045 EXPORT_SYMBOL_GPL(do_exit);
1047 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1055 EXPORT_SYMBOL(complete_and_exit);
1057 SYSCALL_DEFINE1(exit, int, error_code)
1059 do_exit((error_code&0xff)<<8);
1063 * Take down every thread in the group. This is called by fatal signals
1064 * as well as by sys_exit_group (below).
1067 do_group_exit(int exit_code)
1069 struct signal_struct *sig = current->signal;
1071 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1073 if (signal_group_exit(sig))
1074 exit_code = sig->group_exit_code;
1075 else if (!thread_group_empty(current)) {
1076 struct sighand_struct *const sighand = current->sighand;
1077 spin_lock_irq(&sighand->siglock);
1078 if (signal_group_exit(sig))
1079 /* Another thread got here before we took the lock. */
1080 exit_code = sig->group_exit_code;
1082 sig->group_exit_code = exit_code;
1083 sig->flags = SIGNAL_GROUP_EXIT;
1084 zap_other_threads(current);
1086 spin_unlock_irq(&sighand->siglock);
1094 * this kills every thread in the thread group. Note that any externally
1095 * wait4()-ing process will get the correct exit code - even if this
1096 * thread is not the thread group leader.
1098 SYSCALL_DEFINE1(exit_group, int, error_code)
1100 do_group_exit((error_code & 0xff) << 8);
1106 enum pid_type wo_type;
1110 struct siginfo __user *wo_info;
1111 int __user *wo_stat;
1112 struct rusage __user *wo_rusage;
1114 wait_queue_t child_wait;
1119 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1121 if (type != PIDTYPE_PID)
1122 task = task->group_leader;
1123 return task->pids[type].pid;
1126 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1128 return wo->wo_type == PIDTYPE_MAX ||
1129 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1132 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1134 if (!eligible_pid(wo, p))
1136 /* Wait for all children (clone and not) if __WALL is set;
1137 * otherwise, wait for clone children *only* if __WCLONE is
1138 * set; otherwise, wait for non-clone children *only*. (Note:
1139 * A "clone" child here is one that reports to its parent
1140 * using a signal other than SIGCHLD.) */
1141 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1142 && !(wo->wo_flags & __WALL))
1148 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1149 pid_t pid, uid_t uid, int why, int status)
1151 struct siginfo __user *infop;
1152 int retval = wo->wo_rusage
1153 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1156 infop = wo->wo_info;
1159 retval = put_user(SIGCHLD, &infop->si_signo);
1161 retval = put_user(0, &infop->si_errno);
1163 retval = put_user((short)why, &infop->si_code);
1165 retval = put_user(pid, &infop->si_pid);
1167 retval = put_user(uid, &infop->si_uid);
1169 retval = put_user(status, &infop->si_status);
1177 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1178 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1179 * the lock and this task is uninteresting. If we return nonzero, we have
1180 * released the lock and the system call should return.
1182 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1184 unsigned long state;
1185 int retval, status, traced;
1186 pid_t pid = task_pid_vnr(p);
1187 uid_t uid = __task_cred(p)->uid;
1188 struct siginfo __user *infop;
1190 if (!likely(wo->wo_flags & WEXITED))
1193 if (unlikely(wo->wo_flags & WNOWAIT)) {
1194 int exit_code = p->exit_code;
1198 read_unlock(&tasklist_lock);
1199 if ((exit_code & 0x7f) == 0) {
1201 status = exit_code >> 8;
1203 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1204 status = exit_code & 0x7f;
1206 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1210 * Try to move the task's state to DEAD
1211 * only one thread is allowed to do this:
1213 state = xchg(&p->exit_state, EXIT_DEAD);
1214 if (state != EXIT_ZOMBIE) {
1215 BUG_ON(state != EXIT_DEAD);
1219 traced = ptrace_reparented(p);
1221 * It can be ptraced but not reparented, check
1222 * thread_group_leader() to filter out sub-threads.
1224 if (likely(!traced) && thread_group_leader(p)) {
1225 struct signal_struct *psig;
1226 struct signal_struct *sig;
1227 unsigned long maxrss;
1228 cputime_t tgutime, tgstime;
1231 * The resource counters for the group leader are in its
1232 * own task_struct. Those for dead threads in the group
1233 * are in its signal_struct, as are those for the child
1234 * processes it has previously reaped. All these
1235 * accumulate in the parent's signal_struct c* fields.
1237 * We don't bother to take a lock here to protect these
1238 * p->signal fields, because they are only touched by
1239 * __exit_signal, which runs with tasklist_lock
1240 * write-locked anyway, and so is excluded here. We do
1241 * need to protect the access to parent->signal fields,
1242 * as other threads in the parent group can be right
1243 * here reaping other children at the same time.
1245 * We use thread_group_times() to get times for the thread
1246 * group, which consolidates times for all threads in the
1247 * group including the group leader.
1249 thread_group_times(p, &tgutime, &tgstime);
1250 spin_lock_irq(&p->real_parent->sighand->siglock);
1251 psig = p->real_parent->signal;
1254 cputime_add(psig->cutime,
1255 cputime_add(tgutime,
1258 cputime_add(psig->cstime,
1259 cputime_add(tgstime,
1262 cputime_add(psig->cgtime,
1263 cputime_add(p->gtime,
1264 cputime_add(sig->gtime,
1267 p->min_flt + sig->min_flt + sig->cmin_flt;
1269 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1271 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1273 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1275 task_io_get_inblock(p) +
1276 sig->inblock + sig->cinblock;
1278 task_io_get_oublock(p) +
1279 sig->oublock + sig->coublock;
1280 maxrss = max(sig->maxrss, sig->cmaxrss);
1281 if (psig->cmaxrss < maxrss)
1282 psig->cmaxrss = maxrss;
1283 task_io_accounting_add(&psig->ioac, &p->ioac);
1284 task_io_accounting_add(&psig->ioac, &sig->ioac);
1285 spin_unlock_irq(&p->real_parent->sighand->siglock);
1289 * Now we are sure this task is interesting, and no other
1290 * thread can reap it because we set its state to EXIT_DEAD.
1292 read_unlock(&tasklist_lock);
1294 retval = wo->wo_rusage
1295 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1296 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1297 ? p->signal->group_exit_code : p->exit_code;
1298 if (!retval && wo->wo_stat)
1299 retval = put_user(status, wo->wo_stat);
1301 infop = wo->wo_info;
1302 if (!retval && infop)
1303 retval = put_user(SIGCHLD, &infop->si_signo);
1304 if (!retval && infop)
1305 retval = put_user(0, &infop->si_errno);
1306 if (!retval && infop) {
1309 if ((status & 0x7f) == 0) {
1313 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1316 retval = put_user((short)why, &infop->si_code);
1318 retval = put_user(status, &infop->si_status);
1320 if (!retval && infop)
1321 retval = put_user(pid, &infop->si_pid);
1322 if (!retval && infop)
1323 retval = put_user(uid, &infop->si_uid);
1328 write_lock_irq(&tasklist_lock);
1329 /* We dropped tasklist, ptracer could die and untrace */
1332 * If this is not a sub-thread, notify the parent.
1333 * If parent wants a zombie, don't release it now.
1335 if (thread_group_leader(p) &&
1336 !do_notify_parent(p, p->exit_signal)) {
1337 p->exit_state = EXIT_ZOMBIE;
1340 write_unlock_irq(&tasklist_lock);
1348 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1351 if (task_is_stopped_or_traced(p) &&
1352 !(p->jobctl & JOBCTL_LISTENING))
1353 return &p->exit_code;
1355 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1356 return &p->signal->group_exit_code;
1362 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1364 * @ptrace: is the wait for ptrace
1365 * @p: task to wait for
1367 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1370 * read_lock(&tasklist_lock), which is released if return value is
1371 * non-zero. Also, grabs and releases @p->sighand->siglock.
1374 * 0 if wait condition didn't exist and search for other wait conditions
1375 * should continue. Non-zero return, -errno on failure and @p's pid on
1376 * success, implies that tasklist_lock is released and wait condition
1377 * search should terminate.
1379 static int wait_task_stopped(struct wait_opts *wo,
1380 int ptrace, struct task_struct *p)
1382 struct siginfo __user *infop;
1383 int retval, exit_code, *p_code, why;
1384 uid_t uid = 0; /* unneeded, required by compiler */
1388 * Traditionally we see ptrace'd stopped tasks regardless of options.
1390 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1393 if (!task_stopped_code(p, ptrace))
1397 spin_lock_irq(&p->sighand->siglock);
1399 p_code = task_stopped_code(p, ptrace);
1400 if (unlikely(!p_code))
1403 exit_code = *p_code;
1407 if (!unlikely(wo->wo_flags & WNOWAIT))
1412 spin_unlock_irq(&p->sighand->siglock);
1417 * Now we are pretty sure this task is interesting.
1418 * Make sure it doesn't get reaped out from under us while we
1419 * give up the lock and then examine it below. We don't want to
1420 * keep holding onto the tasklist_lock while we call getrusage and
1421 * possibly take page faults for user memory.
1424 pid = task_pid_vnr(p);
1425 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1426 read_unlock(&tasklist_lock);
1428 if (unlikely(wo->wo_flags & WNOWAIT))
1429 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1431 retval = wo->wo_rusage
1432 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1433 if (!retval && wo->wo_stat)
1434 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1436 infop = wo->wo_info;
1437 if (!retval && infop)
1438 retval = put_user(SIGCHLD, &infop->si_signo);
1439 if (!retval && infop)
1440 retval = put_user(0, &infop->si_errno);
1441 if (!retval && infop)
1442 retval = put_user((short)why, &infop->si_code);
1443 if (!retval && infop)
1444 retval = put_user(exit_code, &infop->si_status);
1445 if (!retval && infop)
1446 retval = put_user(pid, &infop->si_pid);
1447 if (!retval && infop)
1448 retval = put_user(uid, &infop->si_uid);
1458 * Handle do_wait work for one task in a live, non-stopped state.
1459 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1460 * the lock and this task is uninteresting. If we return nonzero, we have
1461 * released the lock and the system call should return.
1463 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1469 if (!unlikely(wo->wo_flags & WCONTINUED))
1472 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1475 spin_lock_irq(&p->sighand->siglock);
1476 /* Re-check with the lock held. */
1477 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1478 spin_unlock_irq(&p->sighand->siglock);
1481 if (!unlikely(wo->wo_flags & WNOWAIT))
1482 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1484 spin_unlock_irq(&p->sighand->siglock);
1486 pid = task_pid_vnr(p);
1488 read_unlock(&tasklist_lock);
1491 retval = wo->wo_rusage
1492 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1494 if (!retval && wo->wo_stat)
1495 retval = put_user(0xffff, wo->wo_stat);
1499 retval = wait_noreap_copyout(wo, p, pid, uid,
1500 CLD_CONTINUED, SIGCONT);
1501 BUG_ON(retval == 0);
1508 * Consider @p for a wait by @parent.
1510 * -ECHILD should be in ->notask_error before the first call.
1511 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1512 * Returns zero if the search for a child should continue;
1513 * then ->notask_error is 0 if @p is an eligible child,
1514 * or another error from security_task_wait(), or still -ECHILD.
1516 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1517 struct task_struct *p)
1519 int ret = eligible_child(wo, p);
1523 ret = security_task_wait(p);
1524 if (unlikely(ret < 0)) {
1526 * If we have not yet seen any eligible child,
1527 * then let this error code replace -ECHILD.
1528 * A permission error will give the user a clue
1529 * to look for security policy problems, rather
1530 * than for mysterious wait bugs.
1532 if (wo->notask_error)
1533 wo->notask_error = ret;
1537 /* dead body doesn't have much to contribute */
1538 if (unlikely(p->exit_state == EXIT_DEAD)) {
1540 * But do not ignore this task until the tracer does
1541 * wait_task_zombie()->do_notify_parent().
1543 if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1544 wo->notask_error = 0;
1549 if (p->exit_state == EXIT_ZOMBIE) {
1551 * A zombie ptracee is only visible to its ptracer.
1552 * Notification and reaping will be cascaded to the real
1553 * parent when the ptracer detaches.
1555 if (likely(!ptrace) && unlikely(p->ptrace)) {
1556 /* it will become visible, clear notask_error */
1557 wo->notask_error = 0;
1561 /* we don't reap group leaders with subthreads */
1562 if (!delay_group_leader(p))
1563 return wait_task_zombie(wo, p);
1566 * Allow access to stopped/continued state via zombie by
1567 * falling through. Clearing of notask_error is complex.
1571 * If WEXITED is set, notask_error should naturally be
1572 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1573 * so, if there are live subthreads, there are events to
1574 * wait for. If all subthreads are dead, it's still safe
1575 * to clear - this function will be called again in finite
1576 * amount time once all the subthreads are released and
1577 * will then return without clearing.
1581 * Stopped state is per-task and thus can't change once the
1582 * target task dies. Only continued and exited can happen.
1583 * Clear notask_error if WCONTINUED | WEXITED.
1585 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1586 wo->notask_error = 0;
1589 * If @p is ptraced by a task in its real parent's group,
1590 * hide group stop/continued state when looking at @p as
1591 * the real parent; otherwise, a single stop can be
1592 * reported twice as group and ptrace stops.
1594 * If a ptracer wants to distinguish the two events for its
1595 * own children, it should create a separate process which
1596 * takes the role of real parent.
1598 if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1602 * @p is alive and it's gonna stop, continue or exit, so
1603 * there always is something to wait for.
1605 wo->notask_error = 0;
1609 * Wait for stopped. Depending on @ptrace, different stopped state
1610 * is used and the two don't interact with each other.
1612 ret = wait_task_stopped(wo, ptrace, p);
1617 * Wait for continued. There's only one continued state and the
1618 * ptracer can consume it which can confuse the real parent. Don't
1619 * use WCONTINUED from ptracer. You don't need or want it.
1621 return wait_task_continued(wo, p);
1625 * Do the work of do_wait() for one thread in the group, @tsk.
1627 * -ECHILD should be in ->notask_error before the first call.
1628 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1629 * Returns zero if the search for a child should continue; then
1630 * ->notask_error is 0 if there were any eligible children,
1631 * or another error from security_task_wait(), or still -ECHILD.
1633 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1635 struct task_struct *p;
1637 list_for_each_entry(p, &tsk->children, sibling) {
1638 int ret = wait_consider_task(wo, 0, p);
1646 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1648 struct task_struct *p;
1650 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1651 int ret = wait_consider_task(wo, 1, p);
1659 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1660 int sync, void *key)
1662 struct wait_opts *wo = container_of(wait, struct wait_opts,
1664 struct task_struct *p = key;
1666 if (!eligible_pid(wo, p))
1669 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1672 return default_wake_function(wait, mode, sync, key);
1675 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1677 __wake_up_sync_key(&parent->signal->wait_chldexit,
1678 TASK_INTERRUPTIBLE, 1, p);
1681 static long do_wait(struct wait_opts *wo)
1683 struct task_struct *tsk;
1686 trace_sched_process_wait(wo->wo_pid);
1688 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1689 wo->child_wait.private = current;
1690 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1693 * If there is nothing that can match our critiera just get out.
1694 * We will clear ->notask_error to zero if we see any child that
1695 * might later match our criteria, even if we are not able to reap
1698 wo->notask_error = -ECHILD;
1699 if ((wo->wo_type < PIDTYPE_MAX) &&
1700 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1703 set_current_state(TASK_INTERRUPTIBLE);
1704 read_lock(&tasklist_lock);
1707 retval = do_wait_thread(wo, tsk);
1711 retval = ptrace_do_wait(wo, tsk);
1715 if (wo->wo_flags & __WNOTHREAD)
1717 } while_each_thread(current, tsk);
1718 read_unlock(&tasklist_lock);
1721 retval = wo->notask_error;
1722 if (!retval && !(wo->wo_flags & WNOHANG)) {
1723 retval = -ERESTARTSYS;
1724 if (!signal_pending(current)) {
1730 __set_current_state(TASK_RUNNING);
1731 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1735 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1736 infop, int, options, struct rusage __user *, ru)
1738 struct wait_opts wo;
1739 struct pid *pid = NULL;
1743 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1745 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1758 type = PIDTYPE_PGID;
1766 if (type < PIDTYPE_MAX)
1767 pid = find_get_pid(upid);
1771 wo.wo_flags = options;
1781 * For a WNOHANG return, clear out all the fields
1782 * we would set so the user can easily tell the
1786 ret = put_user(0, &infop->si_signo);
1788 ret = put_user(0, &infop->si_errno);
1790 ret = put_user(0, &infop->si_code);
1792 ret = put_user(0, &infop->si_pid);
1794 ret = put_user(0, &infop->si_uid);
1796 ret = put_user(0, &infop->si_status);
1801 /* avoid REGPARM breakage on x86: */
1802 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1806 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1807 int, options, struct rusage __user *, ru)
1809 struct wait_opts wo;
1810 struct pid *pid = NULL;
1814 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1815 __WNOTHREAD|__WCLONE|__WALL))
1820 else if (upid < 0) {
1821 type = PIDTYPE_PGID;
1822 pid = find_get_pid(-upid);
1823 } else if (upid == 0) {
1824 type = PIDTYPE_PGID;
1825 pid = get_task_pid(current, PIDTYPE_PGID);
1826 } else /* upid > 0 */ {
1828 pid = find_get_pid(upid);
1833 wo.wo_flags = options | WEXITED;
1835 wo.wo_stat = stat_addr;
1840 /* avoid REGPARM breakage on x86: */
1841 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1845 #ifdef __ARCH_WANT_SYS_WAITPID
1848 * sys_waitpid() remains for compatibility. waitpid() should be
1849 * implemented by calling sys_wait4() from libc.a.
1851 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1853 return sys_wait4(pid, stat_addr, options, NULL);