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);
502 EXPORT_SYMBOL_GPL(get_files_struct);
504 void put_files_struct(struct files_struct *files)
508 if (atomic_dec_and_test(&files->count)) {
511 * Free the fd and fdset arrays if we expanded them.
512 * If the fdtable was embedded, pass files for freeing
513 * at the end of the RCU grace period. Otherwise,
514 * you can free files immediately.
517 fdt = files_fdtable(files);
518 if (fdt != &files->fdtab)
519 kmem_cache_free(files_cachep, files);
524 EXPORT_SYMBOL_GPL(put_files_struct);
526 void reset_files_struct(struct files_struct *files)
528 struct task_struct *tsk = current;
529 struct files_struct *old;
535 put_files_struct(old);
538 void exit_files(struct task_struct *tsk)
540 struct files_struct * files = tsk->files;
546 put_files_struct(files);
550 #ifdef CONFIG_MM_OWNER
552 * A task is exiting. If it owned this mm, find a new owner for the mm.
554 void mm_update_next_owner(struct mm_struct *mm)
556 struct task_struct *c, *g, *p = current;
560 * If the exiting or execing task is not the owner, it's
561 * someone else's problem.
566 * The current owner is exiting/execing and there are no other
567 * candidates. Do not leave the mm pointing to a possibly
568 * freed task structure.
570 if (atomic_read(&mm->mm_users) <= 1) {
575 read_lock(&tasklist_lock);
577 * Search in the children
579 list_for_each_entry(c, &p->children, sibling) {
581 goto assign_new_owner;
585 * Search in the siblings
587 list_for_each_entry(c, &p->real_parent->children, sibling) {
589 goto assign_new_owner;
593 * Search through everything else. We should not get
596 do_each_thread(g, c) {
598 goto assign_new_owner;
599 } while_each_thread(g, c);
601 read_unlock(&tasklist_lock);
603 * We found no owner yet mm_users > 1: this implies that we are
604 * most likely racing with swapoff (try_to_unuse()) or /proc or
605 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
614 * The task_lock protects c->mm from changing.
615 * We always want mm->owner->mm == mm
619 * Delay read_unlock() till we have the task_lock()
620 * to ensure that c does not slip away underneath us
622 read_unlock(&tasklist_lock);
632 #endif /* CONFIG_MM_OWNER */
635 * Turn us into a lazy TLB process if we
638 static void exit_mm(struct task_struct * tsk)
640 struct mm_struct *mm = tsk->mm;
641 struct core_state *core_state;
647 * Serialize with any possible pending coredump.
648 * We must hold mmap_sem around checking core_state
649 * and clearing tsk->mm. The core-inducing thread
650 * will increment ->nr_threads for each thread in the
651 * group with ->mm != NULL.
653 down_read(&mm->mmap_sem);
654 core_state = mm->core_state;
656 struct core_thread self;
657 up_read(&mm->mmap_sem);
660 self.next = xchg(&core_state->dumper.next, &self);
662 * Implies mb(), the result of xchg() must be visible
663 * to core_state->dumper.
665 if (atomic_dec_and_test(&core_state->nr_threads))
666 complete(&core_state->startup);
669 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
670 if (!self.task) /* see coredump_finish() */
674 __set_task_state(tsk, TASK_RUNNING);
675 down_read(&mm->mmap_sem);
677 atomic_inc(&mm->mm_count);
678 BUG_ON(mm != tsk->active_mm);
679 /* more a memory barrier than a real lock */
682 up_read(&mm->mmap_sem);
683 enter_lazy_tlb(mm, current);
684 /* We don't want this task to be frozen prematurely */
685 clear_freeze_flag(tsk);
687 mm_update_next_owner(mm);
692 * When we die, we re-parent all our children.
693 * Try to give them to another thread in our thread
694 * group, and if no such member exists, give it to
695 * the child reaper process (ie "init") in our pid
698 static struct task_struct *find_new_reaper(struct task_struct *father)
699 __releases(&tasklist_lock)
700 __acquires(&tasklist_lock)
702 struct pid_namespace *pid_ns = task_active_pid_ns(father);
703 struct task_struct *thread;
706 while_each_thread(father, thread) {
707 if (thread->flags & PF_EXITING)
709 if (unlikely(pid_ns->child_reaper == father))
710 pid_ns->child_reaper = thread;
714 if (unlikely(pid_ns->child_reaper == father)) {
715 write_unlock_irq(&tasklist_lock);
716 if (unlikely(pid_ns == &init_pid_ns))
717 panic("Attempted to kill init!");
719 zap_pid_ns_processes(pid_ns);
720 write_lock_irq(&tasklist_lock);
722 * We can not clear ->child_reaper or leave it alone.
723 * There may by stealth EXIT_DEAD tasks on ->children,
724 * forget_original_parent() must move them somewhere.
726 pid_ns->child_reaper = init_pid_ns.child_reaper;
729 return pid_ns->child_reaper;
733 * Any that need to be release_task'd are put on the @dead list.
735 static void reparent_leader(struct task_struct *father, struct task_struct *p,
736 struct list_head *dead)
738 list_move_tail(&p->sibling, &p->real_parent->children);
740 * If this is a threaded reparent there is no need to
741 * notify anyone anything has happened.
743 if (same_thread_group(p->real_parent, father))
747 * We don't want people slaying init.
749 * Note: we do this even if it is EXIT_DEAD, wait_task_zombie()
750 * can change ->exit_state to EXIT_ZOMBIE. If this is the final
751 * state, do_notify_parent() was already called and ->exit_signal
754 p->exit_signal = SIGCHLD;
756 if (p->exit_state == EXIT_DEAD)
759 /* If it has exited notify the new parent about this child's death. */
761 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
762 if (do_notify_parent(p, p->exit_signal)) {
763 p->exit_state = EXIT_DEAD;
764 list_move_tail(&p->sibling, dead);
768 kill_orphaned_pgrp(p, father);
771 static void forget_original_parent(struct task_struct *father)
773 struct task_struct *p, *n, *reaper;
774 LIST_HEAD(dead_children);
776 write_lock_irq(&tasklist_lock);
778 * Note that exit_ptrace() and find_new_reaper() might
779 * drop tasklist_lock and reacquire it.
782 reaper = find_new_reaper(father);
784 list_for_each_entry_safe(p, n, &father->children, sibling) {
785 struct task_struct *t = p;
787 t->real_parent = reaper;
788 if (t->parent == father) {
790 t->parent = t->real_parent;
792 if (t->pdeath_signal)
793 group_send_sig_info(t->pdeath_signal,
795 } while_each_thread(p, t);
796 reparent_leader(father, p, &dead_children);
798 write_unlock_irq(&tasklist_lock);
800 BUG_ON(!list_empty(&father->children));
802 list_for_each_entry_safe(p, n, &dead_children, sibling) {
803 list_del_init(&p->sibling);
809 * Send signals to all our closest relatives so that they know
810 * to properly mourn us..
812 static void exit_notify(struct task_struct *tsk, int group_dead)
817 * This does two things:
819 * A. Make init inherit all the child processes
820 * B. Check to see if any process groups have become orphaned
821 * as a result of our exiting, and if they have any stopped
822 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
824 forget_original_parent(tsk);
825 exit_task_namespaces(tsk);
827 write_lock_irq(&tasklist_lock);
829 kill_orphaned_pgrp(tsk->group_leader, NULL);
831 if (unlikely(tsk->ptrace)) {
832 int sig = thread_group_leader(tsk) &&
833 thread_group_empty(tsk) &&
834 !ptrace_reparented(tsk) ?
835 tsk->exit_signal : SIGCHLD;
836 autoreap = do_notify_parent(tsk, sig);
837 } else if (thread_group_leader(tsk)) {
838 autoreap = thread_group_empty(tsk) &&
839 do_notify_parent(tsk, tsk->exit_signal);
844 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
846 /* mt-exec, de_thread() is waiting for group leader */
847 if (unlikely(tsk->signal->notify_count < 0))
848 wake_up_process(tsk->signal->group_exit_task);
849 write_unlock_irq(&tasklist_lock);
851 /* If the process is dead, release it - nobody will wait for it */
856 #ifdef CONFIG_DEBUG_STACK_USAGE
857 static void check_stack_usage(void)
859 static DEFINE_SPINLOCK(low_water_lock);
860 static int lowest_to_date = THREAD_SIZE;
863 free = stack_not_used(current);
865 if (free >= lowest_to_date)
868 spin_lock(&low_water_lock);
869 if (free < lowest_to_date) {
870 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
872 current->comm, free);
873 lowest_to_date = free;
875 spin_unlock(&low_water_lock);
878 static inline void check_stack_usage(void) {}
881 NORET_TYPE void do_exit(long code)
883 struct task_struct *tsk = current;
886 profile_task_exit(tsk);
888 WARN_ON(blk_needs_flush_plug(tsk));
890 if (unlikely(in_interrupt()))
891 panic("Aiee, killing interrupt handler!");
892 if (unlikely(!tsk->pid))
893 panic("Attempted to kill the idle task!");
896 * If do_exit is called because this processes oopsed, it's possible
897 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
898 * continuing. Amongst other possible reasons, this is to prevent
899 * mm_release()->clear_child_tid() from writing to a user-controlled
904 ptrace_event(PTRACE_EVENT_EXIT, code);
906 validate_creds_for_do_exit(tsk);
909 * We're taking recursive faults here in do_exit. Safest is to just
910 * leave this task alone and wait for reboot.
912 if (unlikely(tsk->flags & PF_EXITING)) {
914 "Fixing recursive fault but reboot is needed!\n");
916 * We can do this unlocked here. The futex code uses
917 * this flag just to verify whether the pi state
918 * cleanup has been done or not. In the worst case it
919 * loops once more. We pretend that the cleanup was
920 * done as there is no way to return. Either the
921 * OWNER_DIED bit is set by now or we push the blocked
922 * task into the wait for ever nirwana as well.
924 tsk->flags |= PF_EXITPIDONE;
925 set_current_state(TASK_UNINTERRUPTIBLE);
931 exit_signals(tsk); /* sets PF_EXITING */
933 * tsk->flags are checked in the futex code to protect against
934 * an exiting task cleaning up the robust pi futexes.
937 raw_spin_unlock_wait(&tsk->pi_lock);
939 if (unlikely(in_atomic()))
940 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
941 current->comm, task_pid_nr(current),
944 acct_update_integrals(tsk);
945 /* sync mm's RSS info before statistics gathering */
947 sync_mm_rss(tsk, tsk->mm);
948 group_dead = atomic_dec_and_test(&tsk->signal->live);
950 hrtimer_cancel(&tsk->signal->real_timer);
951 exit_itimers(tsk->signal);
953 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
955 acct_collect(code, group_dead);
958 if (unlikely(tsk->audit_context))
961 tsk->exit_code = code;
962 taskstats_exit(tsk, group_dead);
968 trace_sched_process_exit(tsk);
978 * Flush inherited counters to the parent - before the parent
979 * gets woken up by child-exit notifications.
981 * because of cgroup mode, must be called before cgroup_exit()
983 perf_event_exit_task(tsk);
988 disassociate_ctty(1);
990 module_put(task_thread_info(tsk)->exec_domain->module);
992 proc_exit_connector(tsk);
995 * FIXME: do that only when needed, using sched_exit tracepoint
997 ptrace_put_breakpoints(tsk);
999 exit_notify(tsk, group_dead);
1002 mpol_put(tsk->mempolicy);
1003 tsk->mempolicy = NULL;
1007 if (unlikely(current->pi_state_cache))
1008 kfree(current->pi_state_cache);
1011 * Make sure we are holding no locks:
1013 debug_check_no_locks_held(tsk);
1015 * We can do this unlocked here. The futex code uses this flag
1016 * just to verify whether the pi state cleanup has been done
1017 * or not. In the worst case it loops once more.
1019 tsk->flags |= PF_EXITPIDONE;
1021 if (tsk->io_context)
1022 exit_io_context(tsk);
1024 if (tsk->splice_pipe)
1025 __free_pipe_info(tsk->splice_pipe);
1027 validate_creds_for_do_exit(tsk);
1033 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
1034 * when the following two conditions become true.
1035 * - There is race condition of mmap_sem (It is acquired by
1037 * - SMI occurs before setting TASK_RUNINNG.
1038 * (or hypervisor of virtual machine switches to other guest)
1039 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
1041 * To avoid it, we have to wait for releasing tsk->pi_lock which
1042 * is held by try_to_wake_up()
1045 raw_spin_unlock_wait(&tsk->pi_lock);
1047 /* causes final put_task_struct in finish_task_switch(). */
1048 tsk->state = TASK_DEAD;
1051 /* Avoid "noreturn function does return". */
1053 cpu_relax(); /* For when BUG is null */
1056 EXPORT_SYMBOL_GPL(do_exit);
1058 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1066 EXPORT_SYMBOL(complete_and_exit);
1068 SYSCALL_DEFINE1(exit, int, error_code)
1070 do_exit((error_code&0xff)<<8);
1074 * Take down every thread in the group. This is called by fatal signals
1075 * as well as by sys_exit_group (below).
1078 do_group_exit(int exit_code)
1080 struct signal_struct *sig = current->signal;
1082 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1084 if (signal_group_exit(sig))
1085 exit_code = sig->group_exit_code;
1086 else if (!thread_group_empty(current)) {
1087 struct sighand_struct *const sighand = current->sighand;
1088 spin_lock_irq(&sighand->siglock);
1089 if (signal_group_exit(sig))
1090 /* Another thread got here before we took the lock. */
1091 exit_code = sig->group_exit_code;
1093 sig->group_exit_code = exit_code;
1094 sig->flags = SIGNAL_GROUP_EXIT;
1095 zap_other_threads(current);
1097 spin_unlock_irq(&sighand->siglock);
1105 * this kills every thread in the thread group. Note that any externally
1106 * wait4()-ing process will get the correct exit code - even if this
1107 * thread is not the thread group leader.
1109 SYSCALL_DEFINE1(exit_group, int, error_code)
1111 do_group_exit((error_code & 0xff) << 8);
1117 enum pid_type wo_type;
1121 struct siginfo __user *wo_info;
1122 int __user *wo_stat;
1123 struct rusage __user *wo_rusage;
1125 wait_queue_t child_wait;
1130 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1132 if (type != PIDTYPE_PID)
1133 task = task->group_leader;
1134 return task->pids[type].pid;
1137 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1139 return wo->wo_type == PIDTYPE_MAX ||
1140 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1143 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1145 if (!eligible_pid(wo, p))
1147 /* Wait for all children (clone and not) if __WALL is set;
1148 * otherwise, wait for clone children *only* if __WCLONE is
1149 * set; otherwise, wait for non-clone children *only*. (Note:
1150 * A "clone" child here is one that reports to its parent
1151 * using a signal other than SIGCHLD.) */
1152 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1153 && !(wo->wo_flags & __WALL))
1159 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1160 pid_t pid, uid_t uid, int why, int status)
1162 struct siginfo __user *infop;
1163 int retval = wo->wo_rusage
1164 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1167 infop = wo->wo_info;
1170 retval = put_user(SIGCHLD, &infop->si_signo);
1172 retval = put_user(0, &infop->si_errno);
1174 retval = put_user((short)why, &infop->si_code);
1176 retval = put_user(pid, &infop->si_pid);
1178 retval = put_user(uid, &infop->si_uid);
1180 retval = put_user(status, &infop->si_status);
1188 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1189 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1190 * the lock and this task is uninteresting. If we return nonzero, we have
1191 * released the lock and the system call should return.
1193 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1195 unsigned long state;
1196 int retval, status, traced;
1197 pid_t pid = task_pid_vnr(p);
1198 uid_t uid = __task_cred(p)->uid;
1199 struct siginfo __user *infop;
1201 if (!likely(wo->wo_flags & WEXITED))
1204 if (unlikely(wo->wo_flags & WNOWAIT)) {
1205 int exit_code = p->exit_code;
1209 read_unlock(&tasklist_lock);
1210 if ((exit_code & 0x7f) == 0) {
1212 status = exit_code >> 8;
1214 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1215 status = exit_code & 0x7f;
1217 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1221 * Try to move the task's state to DEAD
1222 * only one thread is allowed to do this:
1224 state = xchg(&p->exit_state, EXIT_DEAD);
1225 if (state != EXIT_ZOMBIE) {
1226 BUG_ON(state != EXIT_DEAD);
1230 traced = ptrace_reparented(p);
1232 * It can be ptraced but not reparented, check
1233 * thread_group_leader() to filter out sub-threads.
1235 if (likely(!traced) && thread_group_leader(p)) {
1236 struct signal_struct *psig;
1237 struct signal_struct *sig;
1238 unsigned long maxrss;
1239 cputime_t tgutime, tgstime;
1242 * The resource counters for the group leader are in its
1243 * own task_struct. Those for dead threads in the group
1244 * are in its signal_struct, as are those for the child
1245 * processes it has previously reaped. All these
1246 * accumulate in the parent's signal_struct c* fields.
1248 * We don't bother to take a lock here to protect these
1249 * p->signal fields, because they are only touched by
1250 * __exit_signal, which runs with tasklist_lock
1251 * write-locked anyway, and so is excluded here. We do
1252 * need to protect the access to parent->signal fields,
1253 * as other threads in the parent group can be right
1254 * here reaping other children at the same time.
1256 * We use thread_group_times() to get times for the thread
1257 * group, which consolidates times for all threads in the
1258 * group including the group leader.
1260 thread_group_times(p, &tgutime, &tgstime);
1261 spin_lock_irq(&p->real_parent->sighand->siglock);
1262 psig = p->real_parent->signal;
1265 cputime_add(psig->cutime,
1266 cputime_add(tgutime,
1269 cputime_add(psig->cstime,
1270 cputime_add(tgstime,
1273 cputime_add(psig->cgtime,
1274 cputime_add(p->gtime,
1275 cputime_add(sig->gtime,
1278 p->min_flt + sig->min_flt + sig->cmin_flt;
1280 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1282 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1284 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1286 task_io_get_inblock(p) +
1287 sig->inblock + sig->cinblock;
1289 task_io_get_oublock(p) +
1290 sig->oublock + sig->coublock;
1291 maxrss = max(sig->maxrss, sig->cmaxrss);
1292 if (psig->cmaxrss < maxrss)
1293 psig->cmaxrss = maxrss;
1294 task_io_accounting_add(&psig->ioac, &p->ioac);
1295 task_io_accounting_add(&psig->ioac, &sig->ioac);
1296 spin_unlock_irq(&p->real_parent->sighand->siglock);
1300 * Now we are sure this task is interesting, and no other
1301 * thread can reap it because we set its state to EXIT_DEAD.
1303 read_unlock(&tasklist_lock);
1305 retval = wo->wo_rusage
1306 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1307 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1308 ? p->signal->group_exit_code : p->exit_code;
1309 if (!retval && wo->wo_stat)
1310 retval = put_user(status, wo->wo_stat);
1312 infop = wo->wo_info;
1313 if (!retval && infop)
1314 retval = put_user(SIGCHLD, &infop->si_signo);
1315 if (!retval && infop)
1316 retval = put_user(0, &infop->si_errno);
1317 if (!retval && infop) {
1320 if ((status & 0x7f) == 0) {
1324 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1327 retval = put_user((short)why, &infop->si_code);
1329 retval = put_user(status, &infop->si_status);
1331 if (!retval && infop)
1332 retval = put_user(pid, &infop->si_pid);
1333 if (!retval && infop)
1334 retval = put_user(uid, &infop->si_uid);
1339 write_lock_irq(&tasklist_lock);
1340 /* We dropped tasklist, ptracer could die and untrace */
1343 * If this is not a sub-thread, notify the parent.
1344 * If parent wants a zombie, don't release it now.
1346 if (thread_group_leader(p) &&
1347 !do_notify_parent(p, p->exit_signal)) {
1348 p->exit_state = EXIT_ZOMBIE;
1351 write_unlock_irq(&tasklist_lock);
1359 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1362 if (task_is_stopped_or_traced(p) &&
1363 !(p->jobctl & JOBCTL_LISTENING))
1364 return &p->exit_code;
1366 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1367 return &p->signal->group_exit_code;
1373 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1375 * @ptrace: is the wait for ptrace
1376 * @p: task to wait for
1378 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1381 * read_lock(&tasklist_lock), which is released if return value is
1382 * non-zero. Also, grabs and releases @p->sighand->siglock.
1385 * 0 if wait condition didn't exist and search for other wait conditions
1386 * should continue. Non-zero return, -errno on failure and @p's pid on
1387 * success, implies that tasklist_lock is released and wait condition
1388 * search should terminate.
1390 static int wait_task_stopped(struct wait_opts *wo,
1391 int ptrace, struct task_struct *p)
1393 struct siginfo __user *infop;
1394 int retval, exit_code, *p_code, why;
1395 uid_t uid = 0; /* unneeded, required by compiler */
1399 * Traditionally we see ptrace'd stopped tasks regardless of options.
1401 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1404 if (!task_stopped_code(p, ptrace))
1408 spin_lock_irq(&p->sighand->siglock);
1410 p_code = task_stopped_code(p, ptrace);
1411 if (unlikely(!p_code))
1414 exit_code = *p_code;
1418 if (!unlikely(wo->wo_flags & WNOWAIT))
1423 spin_unlock_irq(&p->sighand->siglock);
1428 * Now we are pretty sure this task is interesting.
1429 * Make sure it doesn't get reaped out from under us while we
1430 * give up the lock and then examine it below. We don't want to
1431 * keep holding onto the tasklist_lock while we call getrusage and
1432 * possibly take page faults for user memory.
1435 pid = task_pid_vnr(p);
1436 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1437 read_unlock(&tasklist_lock);
1439 if (unlikely(wo->wo_flags & WNOWAIT))
1440 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1442 retval = wo->wo_rusage
1443 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1444 if (!retval && wo->wo_stat)
1445 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1447 infop = wo->wo_info;
1448 if (!retval && infop)
1449 retval = put_user(SIGCHLD, &infop->si_signo);
1450 if (!retval && infop)
1451 retval = put_user(0, &infop->si_errno);
1452 if (!retval && infop)
1453 retval = put_user((short)why, &infop->si_code);
1454 if (!retval && infop)
1455 retval = put_user(exit_code, &infop->si_status);
1456 if (!retval && infop)
1457 retval = put_user(pid, &infop->si_pid);
1458 if (!retval && infop)
1459 retval = put_user(uid, &infop->si_uid);
1469 * Handle do_wait work for one task in a live, non-stopped state.
1470 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1471 * the lock and this task is uninteresting. If we return nonzero, we have
1472 * released the lock and the system call should return.
1474 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1480 if (!unlikely(wo->wo_flags & WCONTINUED))
1483 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1486 spin_lock_irq(&p->sighand->siglock);
1487 /* Re-check with the lock held. */
1488 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1489 spin_unlock_irq(&p->sighand->siglock);
1492 if (!unlikely(wo->wo_flags & WNOWAIT))
1493 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1495 spin_unlock_irq(&p->sighand->siglock);
1497 pid = task_pid_vnr(p);
1499 read_unlock(&tasklist_lock);
1502 retval = wo->wo_rusage
1503 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1505 if (!retval && wo->wo_stat)
1506 retval = put_user(0xffff, wo->wo_stat);
1510 retval = wait_noreap_copyout(wo, p, pid, uid,
1511 CLD_CONTINUED, SIGCONT);
1512 BUG_ON(retval == 0);
1519 * Consider @p for a wait by @parent.
1521 * -ECHILD should be in ->notask_error before the first call.
1522 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1523 * Returns zero if the search for a child should continue;
1524 * then ->notask_error is 0 if @p is an eligible child,
1525 * or another error from security_task_wait(), or still -ECHILD.
1527 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1528 struct task_struct *p)
1530 int ret = eligible_child(wo, p);
1534 ret = security_task_wait(p);
1535 if (unlikely(ret < 0)) {
1537 * If we have not yet seen any eligible child,
1538 * then let this error code replace -ECHILD.
1539 * A permission error will give the user a clue
1540 * to look for security policy problems, rather
1541 * than for mysterious wait bugs.
1543 if (wo->notask_error)
1544 wo->notask_error = ret;
1548 /* dead body doesn't have much to contribute */
1549 if (unlikely(p->exit_state == EXIT_DEAD)) {
1551 * But do not ignore this task until the tracer does
1552 * wait_task_zombie()->do_notify_parent().
1554 if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1555 wo->notask_error = 0;
1560 if (p->exit_state == EXIT_ZOMBIE) {
1562 * A zombie ptracee is only visible to its ptracer.
1563 * Notification and reaping will be cascaded to the real
1564 * parent when the ptracer detaches.
1566 if (likely(!ptrace) && unlikely(p->ptrace)) {
1567 /* it will become visible, clear notask_error */
1568 wo->notask_error = 0;
1572 /* we don't reap group leaders with subthreads */
1573 if (!delay_group_leader(p))
1574 return wait_task_zombie(wo, p);
1577 * Allow access to stopped/continued state via zombie by
1578 * falling through. Clearing of notask_error is complex.
1582 * If WEXITED is set, notask_error should naturally be
1583 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1584 * so, if there are live subthreads, there are events to
1585 * wait for. If all subthreads are dead, it's still safe
1586 * to clear - this function will be called again in finite
1587 * amount time once all the subthreads are released and
1588 * will then return without clearing.
1592 * Stopped state is per-task and thus can't change once the
1593 * target task dies. Only continued and exited can happen.
1594 * Clear notask_error if WCONTINUED | WEXITED.
1596 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1597 wo->notask_error = 0;
1600 * If @p is ptraced by a task in its real parent's group,
1601 * hide group stop/continued state when looking at @p as
1602 * the real parent; otherwise, a single stop can be
1603 * reported twice as group and ptrace stops.
1605 * If a ptracer wants to distinguish the two events for its
1606 * own children, it should create a separate process which
1607 * takes the role of real parent.
1609 if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1613 * @p is alive and it's gonna stop, continue or exit, so
1614 * there always is something to wait for.
1616 wo->notask_error = 0;
1620 * Wait for stopped. Depending on @ptrace, different stopped state
1621 * is used and the two don't interact with each other.
1623 ret = wait_task_stopped(wo, ptrace, p);
1628 * Wait for continued. There's only one continued state and the
1629 * ptracer can consume it which can confuse the real parent. Don't
1630 * use WCONTINUED from ptracer. You don't need or want it.
1632 return wait_task_continued(wo, p);
1636 * Do the work of do_wait() for one thread in the group, @tsk.
1638 * -ECHILD should be in ->notask_error before the first call.
1639 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1640 * Returns zero if the search for a child should continue; then
1641 * ->notask_error is 0 if there were any eligible children,
1642 * or another error from security_task_wait(), or still -ECHILD.
1644 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1646 struct task_struct *p;
1648 list_for_each_entry(p, &tsk->children, sibling) {
1649 int ret = wait_consider_task(wo, 0, p);
1657 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1659 struct task_struct *p;
1661 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1662 int ret = wait_consider_task(wo, 1, p);
1670 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1671 int sync, void *key)
1673 struct wait_opts *wo = container_of(wait, struct wait_opts,
1675 struct task_struct *p = key;
1677 if (!eligible_pid(wo, p))
1680 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1683 return default_wake_function(wait, mode, sync, key);
1686 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1688 __wake_up_sync_key(&parent->signal->wait_chldexit,
1689 TASK_INTERRUPTIBLE, 1, p);
1692 static long do_wait(struct wait_opts *wo)
1694 struct task_struct *tsk;
1697 trace_sched_process_wait(wo->wo_pid);
1699 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1700 wo->child_wait.private = current;
1701 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1704 * If there is nothing that can match our critiera just get out.
1705 * We will clear ->notask_error to zero if we see any child that
1706 * might later match our criteria, even if we are not able to reap
1709 wo->notask_error = -ECHILD;
1710 if ((wo->wo_type < PIDTYPE_MAX) &&
1711 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1714 set_current_state(TASK_INTERRUPTIBLE);
1715 read_lock(&tasklist_lock);
1718 retval = do_wait_thread(wo, tsk);
1722 retval = ptrace_do_wait(wo, tsk);
1726 if (wo->wo_flags & __WNOTHREAD)
1728 } while_each_thread(current, tsk);
1729 read_unlock(&tasklist_lock);
1732 retval = wo->notask_error;
1733 if (!retval && !(wo->wo_flags & WNOHANG)) {
1734 retval = -ERESTARTSYS;
1735 if (!signal_pending(current)) {
1741 __set_current_state(TASK_RUNNING);
1742 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1746 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1747 infop, int, options, struct rusage __user *, ru)
1749 struct wait_opts wo;
1750 struct pid *pid = NULL;
1754 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1756 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1769 type = PIDTYPE_PGID;
1777 if (type < PIDTYPE_MAX)
1778 pid = find_get_pid(upid);
1782 wo.wo_flags = options;
1792 * For a WNOHANG return, clear out all the fields
1793 * we would set so the user can easily tell the
1797 ret = put_user(0, &infop->si_signo);
1799 ret = put_user(0, &infop->si_errno);
1801 ret = put_user(0, &infop->si_code);
1803 ret = put_user(0, &infop->si_pid);
1805 ret = put_user(0, &infop->si_uid);
1807 ret = put_user(0, &infop->si_status);
1812 /* avoid REGPARM breakage on x86: */
1813 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1817 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1818 int, options, struct rusage __user *, ru)
1820 struct wait_opts wo;
1821 struct pid *pid = NULL;
1825 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1826 __WNOTHREAD|__WCLONE|__WALL))
1831 else if (upid < 0) {
1832 type = PIDTYPE_PGID;
1833 pid = find_get_pid(-upid);
1834 } else if (upid == 0) {
1835 type = PIDTYPE_PGID;
1836 pid = get_task_pid(current, PIDTYPE_PGID);
1837 } else /* upid > 0 */ {
1839 pid = find_get_pid(upid);
1844 wo.wo_flags = options | WEXITED;
1846 wo.wo_stat = stat_addr;
1851 /* avoid REGPARM breakage on x86: */
1852 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1856 #ifdef __ARCH_WANT_SYS_WAITPID
1859 * sys_waitpid() remains for compatibility. waitpid() should be
1860 * implemented by calling sys_wait4() from libc.a.
1862 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1864 return sys_wait4(pid, stat_addr, options, NULL);