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/mnt_namespace.h>
16 #include <linux/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <linux/init_task.h>
50 #include <trace/sched.h>
52 #include <asm/uaccess.h>
53 #include <asm/unistd.h>
54 #include <asm/pgtable.h>
55 #include <asm/mmu_context.h>
56 #include "cred-internals.h"
58 DEFINE_TRACE(sched_process_free);
59 DEFINE_TRACE(sched_process_exit);
60 DEFINE_TRACE(sched_process_wait);
62 static void exit_mm(struct task_struct * tsk);
64 static void __unhash_process(struct task_struct *p)
67 detach_pid(p, PIDTYPE_PID);
68 if (thread_group_leader(p)) {
69 detach_pid(p, PIDTYPE_PGID);
70 detach_pid(p, PIDTYPE_SID);
72 list_del_rcu(&p->tasks);
73 __get_cpu_var(process_counts)--;
75 list_del_rcu(&p->thread_group);
76 list_del_init(&p->sibling);
80 * This function expects the tasklist_lock write-locked.
82 static void __exit_signal(struct task_struct *tsk)
84 struct signal_struct *sig = tsk->signal;
85 struct sighand_struct *sighand;
88 BUG_ON(!atomic_read(&sig->count));
90 sighand = rcu_dereference(tsk->sighand);
91 spin_lock(&sighand->siglock);
93 posix_cpu_timers_exit(tsk);
94 if (atomic_dec_and_test(&sig->count))
95 posix_cpu_timers_exit_group(tsk);
98 * If there is any task waiting for the group exit
101 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
102 wake_up_process(sig->group_exit_task);
104 if (tsk == sig->curr_target)
105 sig->curr_target = next_thread(tsk);
107 * Accumulate here the counters for all threads but the
108 * group leader as they die, so they can be added into
109 * the process-wide totals when those are taken.
110 * The group leader stays around as a zombie as long
111 * as there are other threads. When it gets reaped,
112 * the exit.c code will add its counts into these totals.
113 * We won't ever get here for the group leader, since it
114 * will have been the last reference on the signal_struct.
116 sig->utime = cputime_add(sig->utime, task_utime(tsk));
117 sig->stime = cputime_add(sig->stime, task_stime(tsk));
118 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
119 sig->min_flt += tsk->min_flt;
120 sig->maj_flt += tsk->maj_flt;
121 sig->nvcsw += tsk->nvcsw;
122 sig->nivcsw += tsk->nivcsw;
123 sig->inblock += task_io_get_inblock(tsk);
124 sig->oublock += task_io_get_oublock(tsk);
125 task_io_accounting_add(&sig->ioac, &tsk->ioac);
126 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
127 sig = NULL; /* Marker for below. */
130 __unhash_process(tsk);
133 * Do this under ->siglock, we can race with another thread
134 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
136 flush_sigqueue(&tsk->pending);
140 spin_unlock(&sighand->siglock);
142 __cleanup_sighand(sighand);
143 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
145 flush_sigqueue(&sig->shared_pending);
146 taskstats_tgid_free(sig);
148 * Make sure ->signal can't go away under rq->lock,
149 * see account_group_exec_runtime().
151 task_rq_unlock_wait(tsk);
152 __cleanup_signal(sig);
156 static void delayed_put_task_struct(struct rcu_head *rhp)
158 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
160 trace_sched_process_free(tsk);
161 put_task_struct(tsk);
165 void release_task(struct task_struct * p)
167 struct task_struct *leader;
170 tracehook_prepare_release_task(p);
171 /* don't need to get the RCU readlock here - the process is dead and
172 * can't be modifying its own credentials */
173 atomic_dec(&__task_cred(p)->user->processes);
176 write_lock_irq(&tasklist_lock);
177 tracehook_finish_release_task(p);
181 * If we are the last non-leader member of the thread
182 * group, and the leader is zombie, then notify the
183 * group leader's parent process. (if it wants notification.)
186 leader = p->group_leader;
187 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
188 BUG_ON(task_detached(leader));
189 do_notify_parent(leader, leader->exit_signal);
191 * If we were the last child thread and the leader has
192 * exited already, and the leader's parent ignores SIGCHLD,
193 * then we are the one who should release the leader.
195 * do_notify_parent() will have marked it self-reaping in
198 zap_leader = task_detached(leader);
201 * This maintains the invariant that release_task()
202 * only runs on a task in EXIT_DEAD, just for sanity.
205 leader->exit_state = EXIT_DEAD;
208 write_unlock_irq(&tasklist_lock);
210 call_rcu(&p->rcu, delayed_put_task_struct);
213 if (unlikely(zap_leader))
218 * This checks not only the pgrp, but falls back on the pid if no
219 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
222 * The caller must hold rcu lock or the tasklist lock.
224 struct pid *session_of_pgrp(struct pid *pgrp)
226 struct task_struct *p;
227 struct pid *sid = NULL;
229 p = pid_task(pgrp, PIDTYPE_PGID);
231 p = pid_task(pgrp, PIDTYPE_PID);
233 sid = task_session(p);
239 * Determine if a process group is "orphaned", according to the POSIX
240 * definition in 2.2.2.52. Orphaned process groups are not to be affected
241 * by terminal-generated stop signals. Newly orphaned process groups are
242 * to receive a SIGHUP and a SIGCONT.
244 * "I ask you, have you ever known what it is to be an orphan?"
246 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
248 struct task_struct *p;
250 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
251 if ((p == ignored_task) ||
252 (p->exit_state && thread_group_empty(p)) ||
253 is_global_init(p->real_parent))
256 if (task_pgrp(p->real_parent) != pgrp &&
257 task_session(p->real_parent) == task_session(p))
259 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
264 int is_current_pgrp_orphaned(void)
268 read_lock(&tasklist_lock);
269 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
270 read_unlock(&tasklist_lock);
275 static int has_stopped_jobs(struct pid *pgrp)
278 struct task_struct *p;
280 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
281 if (!task_is_stopped(p))
285 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
290 * Check to see if any process groups have become orphaned as
291 * a result of our exiting, and if they have any stopped jobs,
292 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
295 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
297 struct pid *pgrp = task_pgrp(tsk);
298 struct task_struct *ignored_task = tsk;
301 /* exit: our father is in a different pgrp than
302 * we are and we were the only connection outside.
304 parent = tsk->real_parent;
306 /* reparent: our child is in a different pgrp than
307 * we are, and it was the only connection outside.
311 if (task_pgrp(parent) != pgrp &&
312 task_session(parent) == task_session(tsk) &&
313 will_become_orphaned_pgrp(pgrp, ignored_task) &&
314 has_stopped_jobs(pgrp)) {
315 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
316 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
321 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
323 * If a kernel thread is launched as a result of a system call, or if
324 * it ever exits, it should generally reparent itself to kthreadd so it
325 * isn't in the way of other processes and is correctly cleaned up on exit.
327 * The various task state such as scheduling policy and priority may have
328 * been inherited from a user process, so we reset them to sane values here.
330 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
332 static void reparent_to_kthreadd(void)
334 write_lock_irq(&tasklist_lock);
336 ptrace_unlink(current);
337 /* Reparent to init */
338 current->real_parent = current->parent = kthreadd_task;
339 list_move_tail(¤t->sibling, ¤t->real_parent->children);
341 /* Set the exit signal to SIGCHLD so we signal init on exit */
342 current->exit_signal = SIGCHLD;
344 if (task_nice(current) < 0)
345 set_user_nice(current, 0);
349 memcpy(current->signal->rlim, init_task.signal->rlim,
350 sizeof(current->signal->rlim));
352 atomic_inc(&init_cred.usage);
353 commit_creds(&init_cred);
354 write_unlock_irq(&tasklist_lock);
357 void __set_special_pids(struct pid *pid)
359 struct task_struct *curr = current->group_leader;
361 if (task_session(curr) != pid)
362 change_pid(curr, PIDTYPE_SID, pid);
364 if (task_pgrp(curr) != pid)
365 change_pid(curr, PIDTYPE_PGID, pid);
368 static void set_special_pids(struct pid *pid)
370 write_lock_irq(&tasklist_lock);
371 __set_special_pids(pid);
372 write_unlock_irq(&tasklist_lock);
376 * Let kernel threads use this to say that they
377 * allow a certain signal (since daemonize() will
378 * have disabled all of them by default).
380 int allow_signal(int sig)
382 if (!valid_signal(sig) || sig < 1)
385 spin_lock_irq(¤t->sighand->siglock);
386 sigdelset(¤t->blocked, sig);
388 /* Kernel threads handle their own signals.
389 Let the signal code know it'll be handled, so
390 that they don't get converted to SIGKILL or
391 just silently dropped */
392 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
395 spin_unlock_irq(¤t->sighand->siglock);
399 EXPORT_SYMBOL(allow_signal);
401 int disallow_signal(int sig)
403 if (!valid_signal(sig) || sig < 1)
406 spin_lock_irq(¤t->sighand->siglock);
407 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
409 spin_unlock_irq(¤t->sighand->siglock);
413 EXPORT_SYMBOL(disallow_signal);
416 * Put all the gunge required to become a kernel thread without
417 * attached user resources in one place where it belongs.
420 void daemonize(const char *name, ...)
423 struct fs_struct *fs;
426 va_start(args, name);
427 vsnprintf(current->comm, sizeof(current->comm), name, args);
431 * If we were started as result of loading a module, close all of the
432 * user space pages. We don't need them, and if we didn't close them
433 * they would be locked into memory.
437 * We don't want to have TIF_FREEZE set if the system-wide hibernation
438 * or suspend transition begins right now.
440 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
442 if (current->nsproxy != &init_nsproxy) {
443 get_nsproxy(&init_nsproxy);
444 switch_task_namespaces(current, &init_nsproxy);
446 set_special_pids(&init_struct_pid);
447 proc_clear_tty(current);
449 /* Block and flush all signals */
450 sigfillset(&blocked);
451 sigprocmask(SIG_BLOCK, &blocked, NULL);
452 flush_signals(current);
454 /* Become as one with the init task */
456 exit_fs(current); /* current->fs->count--; */
459 atomic_inc(&fs->count);
462 current->files = init_task.files;
463 atomic_inc(¤t->files->count);
465 reparent_to_kthreadd();
468 EXPORT_SYMBOL(daemonize);
470 static void close_files(struct files_struct * files)
478 * It is safe to dereference the fd table without RCU or
479 * ->file_lock because this is the last reference to the
482 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.
528 fdt = files_fdtable(files);
529 if (fdt != &files->fdtab)
530 kmem_cache_free(files_cachep, files);
535 void reset_files_struct(struct files_struct *files)
537 struct task_struct *tsk = current;
538 struct files_struct *old;
544 put_files_struct(old);
547 void exit_files(struct task_struct *tsk)
549 struct files_struct * files = tsk->files;
555 put_files_struct(files);
559 void put_fs_struct(struct fs_struct *fs)
561 /* No need to hold fs->lock if we are killing it */
562 if (atomic_dec_and_test(&fs->count)) {
565 kmem_cache_free(fs_cachep, fs);
569 void exit_fs(struct task_struct *tsk)
571 struct fs_struct * fs = tsk->fs;
581 EXPORT_SYMBOL_GPL(exit_fs);
583 #ifdef CONFIG_MM_OWNER
585 * Task p is exiting and it owned mm, lets find a new owner for it
588 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
591 * If there are other users of the mm and the owner (us) is exiting
592 * we need to find a new owner to take on the responsibility.
594 if (atomic_read(&mm->mm_users) <= 1)
601 void mm_update_next_owner(struct mm_struct *mm)
603 struct task_struct *c, *g, *p = current;
606 if (!mm_need_new_owner(mm, p))
609 read_lock(&tasklist_lock);
611 * Search in the children
613 list_for_each_entry(c, &p->children, sibling) {
615 goto assign_new_owner;
619 * Search in the siblings
621 list_for_each_entry(c, &p->parent->children, sibling) {
623 goto assign_new_owner;
627 * Search through everything else. We should not get
630 do_each_thread(g, c) {
632 goto assign_new_owner;
633 } while_each_thread(g, c);
635 read_unlock(&tasklist_lock);
637 * We found no owner yet mm_users > 1: this implies that we are
638 * most likely racing with swapoff (try_to_unuse()) or /proc or
639 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
648 * The task_lock protects c->mm from changing.
649 * We always want mm->owner->mm == mm
653 * Delay read_unlock() till we have the task_lock()
654 * to ensure that c does not slip away underneath us
656 read_unlock(&tasklist_lock);
666 #endif /* CONFIG_MM_OWNER */
669 * Turn us into a lazy TLB process if we
672 static void exit_mm(struct task_struct * tsk)
674 struct mm_struct *mm = tsk->mm;
675 struct core_state *core_state;
681 * Serialize with any possible pending coredump.
682 * We must hold mmap_sem around checking core_state
683 * and clearing tsk->mm. The core-inducing thread
684 * will increment ->nr_threads for each thread in the
685 * group with ->mm != NULL.
687 down_read(&mm->mmap_sem);
688 core_state = mm->core_state;
690 struct core_thread self;
691 up_read(&mm->mmap_sem);
694 self.next = xchg(&core_state->dumper.next, &self);
696 * Implies mb(), the result of xchg() must be visible
697 * to core_state->dumper.
699 if (atomic_dec_and_test(&core_state->nr_threads))
700 complete(&core_state->startup);
703 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
704 if (!self.task) /* see coredump_finish() */
708 __set_task_state(tsk, TASK_RUNNING);
709 down_read(&mm->mmap_sem);
711 atomic_inc(&mm->mm_count);
712 BUG_ON(mm != tsk->active_mm);
713 /* more a memory barrier than a real lock */
716 up_read(&mm->mmap_sem);
717 enter_lazy_tlb(mm, current);
718 /* We don't want this task to be frozen prematurely */
719 clear_freeze_flag(tsk);
721 mm_update_next_owner(mm);
726 * When we die, we re-parent all our children.
727 * Try to give them to another thread in our thread
728 * group, and if no such member exists, give it to
729 * the child reaper process (ie "init") in our pid
732 static struct task_struct *find_new_reaper(struct task_struct *father)
734 struct pid_namespace *pid_ns = task_active_pid_ns(father);
735 struct task_struct *thread;
738 while_each_thread(father, thread) {
739 if (thread->flags & PF_EXITING)
741 if (unlikely(pid_ns->child_reaper == father))
742 pid_ns->child_reaper = thread;
746 if (unlikely(pid_ns->child_reaper == father)) {
747 write_unlock_irq(&tasklist_lock);
748 if (unlikely(pid_ns == &init_pid_ns))
749 panic("Attempted to kill init!");
751 zap_pid_ns_processes(pid_ns);
752 write_lock_irq(&tasklist_lock);
754 * We can not clear ->child_reaper or leave it alone.
755 * There may by stealth EXIT_DEAD tasks on ->children,
756 * forget_original_parent() must move them somewhere.
758 pid_ns->child_reaper = init_pid_ns.child_reaper;
761 return pid_ns->child_reaper;
765 * Any that need to be release_task'd are put on the @dead list.
767 static void reparent_thread(struct task_struct *father, struct task_struct *p,
768 struct list_head *dead)
770 if (p->pdeath_signal)
771 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
773 list_move_tail(&p->sibling, &p->real_parent->children);
775 if (task_detached(p))
778 * If this is a threaded reparent there is no need to
779 * notify anyone anything has happened.
781 if (same_thread_group(p->real_parent, father))
784 /* We don't want people slaying init. */
785 p->exit_signal = SIGCHLD;
787 /* If it has exited notify the new parent about this child's death. */
789 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
790 do_notify_parent(p, p->exit_signal);
791 if (task_detached(p)) {
792 p->exit_state = EXIT_DEAD;
793 list_move_tail(&p->sibling, dead);
797 kill_orphaned_pgrp(p, father);
800 static void forget_original_parent(struct task_struct *father)
802 struct task_struct *p, *n, *reaper;
803 LIST_HEAD(dead_children);
807 write_lock_irq(&tasklist_lock);
808 reaper = find_new_reaper(father);
810 list_for_each_entry_safe(p, n, &father->children, sibling) {
811 p->real_parent = reaper;
812 if (p->parent == father) {
814 p->parent = p->real_parent;
816 reparent_thread(father, p, &dead_children);
818 write_unlock_irq(&tasklist_lock);
820 BUG_ON(!list_empty(&father->children));
822 list_for_each_entry_safe(p, n, &dead_children, sibling) {
823 list_del_init(&p->sibling);
829 * Send signals to all our closest relatives so that they know
830 * to properly mourn us..
832 static void exit_notify(struct task_struct *tsk, int group_dead)
838 * This does two things:
840 * A. Make init inherit all the child processes
841 * B. Check to see if any process groups have become orphaned
842 * as a result of our exiting, and if they have any stopped
843 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
845 forget_original_parent(tsk);
846 exit_task_namespaces(tsk);
848 write_lock_irq(&tasklist_lock);
850 kill_orphaned_pgrp(tsk->group_leader, NULL);
852 /* Let father know we died
854 * Thread signals are configurable, but you aren't going to use
855 * that to send signals to arbitary processes.
856 * That stops right now.
858 * If the parent exec id doesn't match the exec id we saved
859 * when we started then we know the parent has changed security
862 * If our self_exec id doesn't match our parent_exec_id then
863 * we have changed execution domain as these two values started
864 * the same after a fork.
866 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
867 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
868 tsk->self_exec_id != tsk->parent_exec_id) &&
870 tsk->exit_signal = SIGCHLD;
872 signal = tracehook_notify_death(tsk, &cookie, group_dead);
874 signal = do_notify_parent(tsk, signal);
876 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
878 /* mt-exec, de_thread() is waiting for us */
879 if (thread_group_leader(tsk) &&
880 tsk->signal->group_exit_task &&
881 tsk->signal->notify_count < 0)
882 wake_up_process(tsk->signal->group_exit_task);
884 write_unlock_irq(&tasklist_lock);
886 tracehook_report_death(tsk, signal, cookie, group_dead);
888 /* If the process is dead, release it - nobody will wait for it */
889 if (signal == DEATH_REAP)
893 #ifdef CONFIG_DEBUG_STACK_USAGE
894 static void check_stack_usage(void)
896 static DEFINE_SPINLOCK(low_water_lock);
897 static int lowest_to_date = THREAD_SIZE;
900 free = stack_not_used(current);
902 if (free >= lowest_to_date)
905 spin_lock(&low_water_lock);
906 if (free < lowest_to_date) {
907 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
909 current->comm, free);
910 lowest_to_date = free;
912 spin_unlock(&low_water_lock);
915 static inline void check_stack_usage(void) {}
918 NORET_TYPE void do_exit(long code)
920 struct task_struct *tsk = current;
923 profile_task_exit(tsk);
925 WARN_ON(atomic_read(&tsk->fs_excl));
927 if (unlikely(in_interrupt()))
928 panic("Aiee, killing interrupt handler!");
929 if (unlikely(!tsk->pid))
930 panic("Attempted to kill the idle task!");
932 tracehook_report_exit(&code);
935 * We're taking recursive faults here in do_exit. Safest is to just
936 * leave this task alone and wait for reboot.
938 if (unlikely(tsk->flags & PF_EXITING)) {
940 "Fixing recursive fault but reboot is needed!\n");
942 * We can do this unlocked here. The futex code uses
943 * this flag just to verify whether the pi state
944 * cleanup has been done or not. In the worst case it
945 * loops once more. We pretend that the cleanup was
946 * done as there is no way to return. Either the
947 * OWNER_DIED bit is set by now or we push the blocked
948 * task into the wait for ever nirwana as well.
950 tsk->flags |= PF_EXITPIDONE;
951 set_current_state(TASK_UNINTERRUPTIBLE);
955 exit_signals(tsk); /* sets PF_EXITING */
957 * tsk->flags are checked in the futex code to protect against
958 * an exiting task cleaning up the robust pi futexes.
961 spin_unlock_wait(&tsk->pi_lock);
963 if (unlikely(in_atomic()))
964 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
965 current->comm, task_pid_nr(current),
968 acct_update_integrals(tsk);
970 group_dead = atomic_dec_and_test(&tsk->signal->live);
972 hrtimer_cancel(&tsk->signal->real_timer);
973 exit_itimers(tsk->signal);
975 acct_collect(code, group_dead);
978 if (unlikely(tsk->audit_context))
981 tsk->exit_code = code;
982 taskstats_exit(tsk, group_dead);
988 trace_sched_process_exit(tsk);
997 if (group_dead && tsk->signal->leader)
998 disassociate_ctty(1);
1000 module_put(task_thread_info(tsk)->exec_domain->module);
1002 module_put(tsk->binfmt->module);
1004 proc_exit_connector(tsk);
1005 exit_notify(tsk, group_dead);
1007 mpol_put(tsk->mempolicy);
1008 tsk->mempolicy = NULL;
1012 * This must happen late, after the PID is not
1015 if (unlikely(!list_empty(&tsk->pi_state_list)))
1016 exit_pi_state_list(tsk);
1017 if (unlikely(current->pi_state_cache))
1018 kfree(current->pi_state_cache);
1021 * Make sure we are holding no locks:
1023 debug_check_no_locks_held(tsk);
1025 * We can do this unlocked here. The futex code uses this flag
1026 * just to verify whether the pi state cleanup has been done
1027 * or not. In the worst case it loops once more.
1029 tsk->flags |= PF_EXITPIDONE;
1031 if (tsk->io_context)
1034 if (tsk->splice_pipe)
1035 __free_pipe_info(tsk->splice_pipe);
1038 /* causes final put_task_struct in finish_task_switch(). */
1039 tsk->state = TASK_DEAD;
1042 /* Avoid "noreturn function does return". */
1044 cpu_relax(); /* For when BUG is null */
1047 EXPORT_SYMBOL_GPL(do_exit);
1049 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1057 EXPORT_SYMBOL(complete_and_exit);
1059 SYSCALL_DEFINE1(exit, int, error_code)
1061 do_exit((error_code&0xff)<<8);
1065 * Take down every thread in the group. This is called by fatal signals
1066 * as well as by sys_exit_group (below).
1069 do_group_exit(int exit_code)
1071 struct signal_struct *sig = current->signal;
1073 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1075 if (signal_group_exit(sig))
1076 exit_code = sig->group_exit_code;
1077 else if (!thread_group_empty(current)) {
1078 struct sighand_struct *const sighand = current->sighand;
1079 spin_lock_irq(&sighand->siglock);
1080 if (signal_group_exit(sig))
1081 /* Another thread got here before we took the lock. */
1082 exit_code = sig->group_exit_code;
1084 sig->group_exit_code = exit_code;
1085 sig->flags = SIGNAL_GROUP_EXIT;
1086 zap_other_threads(current);
1088 spin_unlock_irq(&sighand->siglock);
1096 * this kills every thread in the thread group. Note that any externally
1097 * wait4()-ing process will get the correct exit code - even if this
1098 * thread is not the thread group leader.
1100 SYSCALL_DEFINE1(exit_group, int, error_code)
1102 do_group_exit((error_code & 0xff) << 8);
1107 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1109 struct pid *pid = NULL;
1110 if (type == PIDTYPE_PID)
1111 pid = task->pids[type].pid;
1112 else if (type < PIDTYPE_MAX)
1113 pid = task->group_leader->pids[type].pid;
1117 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1118 struct task_struct *p)
1122 if (type < PIDTYPE_MAX) {
1123 if (task_pid_type(p, type) != pid)
1127 /* Wait for all children (clone and not) if __WALL is set;
1128 * otherwise, wait for clone children *only* if __WCLONE is
1129 * set; otherwise, wait for non-clone children *only*. (Note:
1130 * A "clone" child here is one that reports to its parent
1131 * using a signal other than SIGCHLD.) */
1132 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1133 && !(options & __WALL))
1136 err = security_task_wait(p);
1143 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1144 int why, int status,
1145 struct siginfo __user *infop,
1146 struct rusage __user *rusagep)
1148 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1152 retval = put_user(SIGCHLD, &infop->si_signo);
1154 retval = put_user(0, &infop->si_errno);
1156 retval = put_user((short)why, &infop->si_code);
1158 retval = put_user(pid, &infop->si_pid);
1160 retval = put_user(uid, &infop->si_uid);
1162 retval = put_user(status, &infop->si_status);
1169 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1170 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1171 * the lock and this task is uninteresting. If we return nonzero, we have
1172 * released the lock and the system call should return.
1174 static int wait_task_zombie(struct task_struct *p, int options,
1175 struct siginfo __user *infop,
1176 int __user *stat_addr, struct rusage __user *ru)
1178 unsigned long state;
1179 int retval, status, traced;
1180 pid_t pid = task_pid_vnr(p);
1181 uid_t uid = __task_cred(p)->uid;
1183 if (!likely(options & WEXITED))
1186 if (unlikely(options & WNOWAIT)) {
1187 int exit_code = p->exit_code;
1191 read_unlock(&tasklist_lock);
1192 if ((exit_code & 0x7f) == 0) {
1194 status = exit_code >> 8;
1196 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1197 status = exit_code & 0x7f;
1199 return wait_noreap_copyout(p, pid, uid, why,
1204 * Try to move the task's state to DEAD
1205 * only one thread is allowed to do this:
1207 state = xchg(&p->exit_state, EXIT_DEAD);
1208 if (state != EXIT_ZOMBIE) {
1209 BUG_ON(state != EXIT_DEAD);
1213 traced = ptrace_reparented(p);
1215 if (likely(!traced)) {
1216 struct signal_struct *psig;
1217 struct signal_struct *sig;
1218 struct task_cputime cputime;
1221 * The resource counters for the group leader are in its
1222 * own task_struct. Those for dead threads in the group
1223 * are in its signal_struct, as are those for the child
1224 * processes it has previously reaped. All these
1225 * accumulate in the parent's signal_struct c* fields.
1227 * We don't bother to take a lock here to protect these
1228 * p->signal fields, because they are only touched by
1229 * __exit_signal, which runs with tasklist_lock
1230 * write-locked anyway, and so is excluded here. We do
1231 * need to protect the access to p->parent->signal fields,
1232 * as other threads in the parent group can be right
1233 * here reaping other children at the same time.
1235 * We use thread_group_cputime() to get times for the thread
1236 * group, which consolidates times for all threads in the
1237 * group including the group leader.
1239 thread_group_cputime(p, &cputime);
1240 spin_lock_irq(&p->parent->sighand->siglock);
1241 psig = p->parent->signal;
1244 cputime_add(psig->cutime,
1245 cputime_add(cputime.utime,
1248 cputime_add(psig->cstime,
1249 cputime_add(cputime.stime,
1252 cputime_add(psig->cgtime,
1253 cputime_add(p->gtime,
1254 cputime_add(sig->gtime,
1257 p->min_flt + sig->min_flt + sig->cmin_flt;
1259 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1261 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1263 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1265 task_io_get_inblock(p) +
1266 sig->inblock + sig->cinblock;
1268 task_io_get_oublock(p) +
1269 sig->oublock + sig->coublock;
1270 task_io_accounting_add(&psig->ioac, &p->ioac);
1271 task_io_accounting_add(&psig->ioac, &sig->ioac);
1272 spin_unlock_irq(&p->parent->sighand->siglock);
1276 * Now we are sure this task is interesting, and no other
1277 * thread can reap it because we set its state to EXIT_DEAD.
1279 read_unlock(&tasklist_lock);
1281 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1282 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1283 ? p->signal->group_exit_code : p->exit_code;
1284 if (!retval && stat_addr)
1285 retval = put_user(status, stat_addr);
1286 if (!retval && infop)
1287 retval = put_user(SIGCHLD, &infop->si_signo);
1288 if (!retval && infop)
1289 retval = put_user(0, &infop->si_errno);
1290 if (!retval && infop) {
1293 if ((status & 0x7f) == 0) {
1297 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1300 retval = put_user((short)why, &infop->si_code);
1302 retval = put_user(status, &infop->si_status);
1304 if (!retval && infop)
1305 retval = put_user(pid, &infop->si_pid);
1306 if (!retval && infop)
1307 retval = put_user(uid, &infop->si_uid);
1312 write_lock_irq(&tasklist_lock);
1313 /* We dropped tasklist, ptracer could die and untrace */
1316 * If this is not a detached task, notify the parent.
1317 * If it's still not detached after that, don't release
1320 if (!task_detached(p)) {
1321 do_notify_parent(p, p->exit_signal);
1322 if (!task_detached(p)) {
1323 p->exit_state = EXIT_ZOMBIE;
1327 write_unlock_irq(&tasklist_lock);
1335 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1338 if (task_is_stopped_or_traced(p))
1339 return &p->exit_code;
1341 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1342 return &p->signal->group_exit_code;
1348 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1349 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1350 * the lock and this task is uninteresting. If we return nonzero, we have
1351 * released the lock and the system call should return.
1353 static int wait_task_stopped(int ptrace, struct task_struct *p,
1354 int options, struct siginfo __user *infop,
1355 int __user *stat_addr, struct rusage __user *ru)
1357 int retval, exit_code, *p_code, why;
1358 uid_t uid = 0; /* unneeded, required by compiler */
1361 if (!(options & WUNTRACED))
1365 spin_lock_irq(&p->sighand->siglock);
1367 p_code = task_stopped_code(p, ptrace);
1368 if (unlikely(!p_code))
1371 exit_code = *p_code;
1375 if (!unlikely(options & WNOWAIT))
1378 /* don't need the RCU readlock here as we're holding a spinlock */
1379 uid = __task_cred(p)->uid;
1381 spin_unlock_irq(&p->sighand->siglock);
1386 * Now we are pretty sure this task is interesting.
1387 * Make sure it doesn't get reaped out from under us while we
1388 * give up the lock and then examine it below. We don't want to
1389 * keep holding onto the tasklist_lock while we call getrusage and
1390 * possibly take page faults for user memory.
1393 pid = task_pid_vnr(p);
1394 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1395 read_unlock(&tasklist_lock);
1397 if (unlikely(options & WNOWAIT))
1398 return wait_noreap_copyout(p, pid, uid,
1402 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1403 if (!retval && stat_addr)
1404 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1405 if (!retval && infop)
1406 retval = put_user(SIGCHLD, &infop->si_signo);
1407 if (!retval && infop)
1408 retval = put_user(0, &infop->si_errno);
1409 if (!retval && infop)
1410 retval = put_user((short)why, &infop->si_code);
1411 if (!retval && infop)
1412 retval = put_user(exit_code, &infop->si_status);
1413 if (!retval && infop)
1414 retval = put_user(pid, &infop->si_pid);
1415 if (!retval && infop)
1416 retval = put_user(uid, &infop->si_uid);
1426 * Handle do_wait work for one task in a live, non-stopped state.
1427 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1428 * the lock and this task is uninteresting. If we return nonzero, we have
1429 * released the lock and the system call should return.
1431 static int wait_task_continued(struct task_struct *p, int options,
1432 struct siginfo __user *infop,
1433 int __user *stat_addr, struct rusage __user *ru)
1439 if (!unlikely(options & WCONTINUED))
1442 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1445 spin_lock_irq(&p->sighand->siglock);
1446 /* Re-check with the lock held. */
1447 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1448 spin_unlock_irq(&p->sighand->siglock);
1451 if (!unlikely(options & WNOWAIT))
1452 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1453 uid = __task_cred(p)->uid;
1454 spin_unlock_irq(&p->sighand->siglock);
1456 pid = task_pid_vnr(p);
1458 read_unlock(&tasklist_lock);
1461 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1463 if (!retval && stat_addr)
1464 retval = put_user(0xffff, stat_addr);
1468 retval = wait_noreap_copyout(p, pid, uid,
1469 CLD_CONTINUED, SIGCONT,
1471 BUG_ON(retval == 0);
1478 * Consider @p for a wait by @parent.
1480 * -ECHILD should be in *@notask_error before the first call.
1481 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1482 * Returns zero if the search for a child should continue;
1483 * then *@notask_error is 0 if @p is an eligible child,
1484 * or another error from security_task_wait(), or still -ECHILD.
1486 static int wait_consider_task(struct task_struct *parent, int ptrace,
1487 struct task_struct *p, int *notask_error,
1488 enum pid_type type, struct pid *pid, int options,
1489 struct siginfo __user *infop,
1490 int __user *stat_addr, struct rusage __user *ru)
1492 int ret = eligible_child(type, pid, options, p);
1496 if (unlikely(ret < 0)) {
1498 * If we have not yet seen any eligible child,
1499 * then let this error code replace -ECHILD.
1500 * A permission error will give the user a clue
1501 * to look for security policy problems, rather
1502 * than for mysterious wait bugs.
1505 *notask_error = ret;
1508 if (likely(!ptrace) && unlikely(p->ptrace)) {
1510 * This child is hidden by ptrace.
1511 * We aren't allowed to see it now, but eventually we will.
1517 if (p->exit_state == EXIT_DEAD)
1521 * We don't reap group leaders with subthreads.
1523 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1524 return wait_task_zombie(p, options, infop, stat_addr, ru);
1527 * It's stopped or running now, so it might
1528 * later continue, exit, or stop again.
1532 if (task_stopped_code(p, ptrace))
1533 return wait_task_stopped(ptrace, p, options,
1534 infop, stat_addr, ru);
1536 return wait_task_continued(p, options, infop, stat_addr, ru);
1540 * Do the work of do_wait() for one thread in the group, @tsk.
1542 * -ECHILD should be in *@notask_error before the first call.
1543 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1544 * Returns zero if the search for a child should continue; then
1545 * *@notask_error is 0 if there were any eligible children,
1546 * or another error from security_task_wait(), or still -ECHILD.
1548 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1549 enum pid_type type, struct pid *pid, int options,
1550 struct siginfo __user *infop, int __user *stat_addr,
1551 struct rusage __user *ru)
1553 struct task_struct *p;
1555 list_for_each_entry(p, &tsk->children, sibling) {
1557 * Do not consider detached threads.
1559 if (!task_detached(p)) {
1560 int ret = wait_consider_task(tsk, 0, p, notask_error,
1562 infop, stat_addr, ru);
1571 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1572 enum pid_type type, struct pid *pid, int options,
1573 struct siginfo __user *infop, int __user *stat_addr,
1574 struct rusage __user *ru)
1576 struct task_struct *p;
1579 * Traditionally we see ptrace'd stopped tasks regardless of options.
1581 options |= WUNTRACED;
1583 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1584 int ret = wait_consider_task(tsk, 1, p, notask_error,
1586 infop, stat_addr, ru);
1594 static long do_wait(enum pid_type type, struct pid *pid, int options,
1595 struct siginfo __user *infop, int __user *stat_addr,
1596 struct rusage __user *ru)
1598 DECLARE_WAITQUEUE(wait, current);
1599 struct task_struct *tsk;
1602 trace_sched_process_wait(pid);
1604 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1607 * If there is nothing that can match our critiera just get out.
1608 * We will clear @retval to zero if we see any child that might later
1609 * match our criteria, even if we are not able to reap it yet.
1612 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1615 current->state = TASK_INTERRUPTIBLE;
1616 read_lock(&tasklist_lock);
1619 int tsk_result = do_wait_thread(tsk, &retval,
1621 infop, stat_addr, ru);
1623 tsk_result = ptrace_do_wait(tsk, &retval,
1625 infop, stat_addr, ru);
1628 * tasklist_lock is unlocked and we have a final result.
1630 retval = tsk_result;
1634 if (options & __WNOTHREAD)
1636 tsk = next_thread(tsk);
1637 BUG_ON(tsk->signal != current->signal);
1638 } while (tsk != current);
1639 read_unlock(&tasklist_lock);
1641 if (!retval && !(options & WNOHANG)) {
1642 retval = -ERESTARTSYS;
1643 if (!signal_pending(current)) {
1650 current->state = TASK_RUNNING;
1651 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1657 * For a WNOHANG return, clear out all the fields
1658 * we would set so the user can easily tell the
1662 retval = put_user(0, &infop->si_signo);
1664 retval = put_user(0, &infop->si_errno);
1666 retval = put_user(0, &infop->si_code);
1668 retval = put_user(0, &infop->si_pid);
1670 retval = put_user(0, &infop->si_uid);
1672 retval = put_user(0, &infop->si_status);
1678 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1679 infop, int, options, struct rusage __user *, ru)
1681 struct pid *pid = NULL;
1685 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1687 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1700 type = PIDTYPE_PGID;
1708 if (type < PIDTYPE_MAX)
1709 pid = find_get_pid(upid);
1710 ret = do_wait(type, pid, options, infop, NULL, ru);
1713 /* avoid REGPARM breakage on x86: */
1714 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1718 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1719 int, options, struct rusage __user *, ru)
1721 struct pid *pid = NULL;
1725 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1726 __WNOTHREAD|__WCLONE|__WALL))
1731 else if (upid < 0) {
1732 type = PIDTYPE_PGID;
1733 pid = find_get_pid(-upid);
1734 } else if (upid == 0) {
1735 type = PIDTYPE_PGID;
1736 pid = get_task_pid(current, PIDTYPE_PGID);
1737 } else /* upid > 0 */ {
1739 pid = find_get_pid(upid);
1742 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1745 /* avoid REGPARM breakage on x86: */
1746 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1750 #ifdef __ARCH_WANT_SYS_WAITPID
1753 * sys_waitpid() remains for compatibility. waitpid() should be
1754 * implemented by calling sys_wait4() from libc.a.
1756 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1758 return sys_wait4(pid, stat_addr, options, NULL);