4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/iocontext.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/fdtable.h>
23 #include <linux/binfmts.h>
24 #include <linux/nsproxy.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/ptrace.h>
27 #include <linux/profile.h>
28 #include <linux/mount.h>
29 #include <linux/proc_fs.h>
30 #include <linux/kthread.h>
31 #include <linux/mempolicy.h>
32 #include <linux/taskstats_kern.h>
33 #include <linux/delayacct.h>
34 #include <linux/freezer.h>
35 #include <linux/cgroup.h>
36 #include <linux/syscalls.h>
37 #include <linux/signal.h>
38 #include <linux/posix-timers.h>
39 #include <linux/cn_proc.h>
40 #include <linux/mutex.h>
41 #include <linux/futex.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/tracehook.h>
48 #include <linux/fs_struct.h>
49 #include <linux/init_task.h>
50 #include <linux/perf_event.h>
51 #include <trace/events/sched.h>
52 #include <linux/hw_breakpoint.h>
53 #include <linux/oom.h>
55 #include <asm/uaccess.h>
56 #include <asm/unistd.h>
57 #include <asm/pgtable.h>
58 #include <asm/mmu_context.h>
60 static void exit_mm(struct task_struct * tsk);
62 static void __unhash_process(struct task_struct *p, bool group_dead)
65 detach_pid(p, PIDTYPE_PID);
67 detach_pid(p, PIDTYPE_PGID);
68 detach_pid(p, PIDTYPE_SID);
70 list_del_rcu(&p->tasks);
71 list_del_init(&p->sibling);
72 __this_cpu_dec(process_counts);
74 list_del_rcu(&p->thread_group);
78 * This function expects the tasklist_lock write-locked.
80 static void __exit_signal(struct task_struct *tsk)
82 struct signal_struct *sig = tsk->signal;
83 bool group_dead = thread_group_leader(tsk);
84 struct sighand_struct *sighand;
85 struct tty_struct *uninitialized_var(tty);
87 sighand = rcu_dereference_check(tsk->sighand,
88 rcu_read_lock_held() ||
89 lockdep_tasklist_lock_is_held());
90 spin_lock(&sighand->siglock);
92 posix_cpu_timers_exit(tsk);
94 posix_cpu_timers_exit_group(tsk);
99 * This can only happen if the caller is de_thread().
100 * FIXME: this is the temporary hack, we should teach
101 * posix-cpu-timers to handle this case correctly.
103 if (unlikely(has_group_leader_pid(tsk)))
104 posix_cpu_timers_exit_group(tsk);
107 * If there is any task waiting for the group exit
110 if (sig->notify_count > 0 && !--sig->notify_count)
111 wake_up_process(sig->group_exit_task);
113 if (tsk == sig->curr_target)
114 sig->curr_target = next_thread(tsk);
116 * Accumulate here the counters for all threads but the
117 * group leader as they die, so they can be added into
118 * the process-wide totals when those are taken.
119 * The group leader stays around as a zombie as long
120 * as there are other threads. When it gets reaped,
121 * the exit.c code will add its counts into these totals.
122 * We won't ever get here for the group leader, since it
123 * will have been the last reference on the signal_struct.
125 sig->utime = cputime_add(sig->utime, tsk->utime);
126 sig->stime = cputime_add(sig->stime, tsk->stime);
127 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
128 sig->min_flt += tsk->min_flt;
129 sig->maj_flt += tsk->maj_flt;
130 sig->nvcsw += tsk->nvcsw;
131 sig->nivcsw += tsk->nivcsw;
132 sig->inblock += task_io_get_inblock(tsk);
133 sig->oublock += task_io_get_oublock(tsk);
134 task_io_accounting_add(&sig->ioac, &tsk->ioac);
135 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
139 __unhash_process(tsk, group_dead);
142 * Do this under ->siglock, we can race with another thread
143 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
145 flush_sigqueue(&tsk->pending);
147 spin_unlock(&sighand->siglock);
149 __cleanup_sighand(sighand);
150 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
152 flush_sigqueue(&sig->shared_pending);
157 static void delayed_put_task_struct(struct rcu_head *rhp)
159 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
161 perf_event_delayed_put(tsk);
162 trace_sched_process_free(tsk);
163 put_task_struct(tsk);
167 void release_task(struct task_struct * p)
169 struct task_struct *leader;
172 /* don't need to get the RCU readlock here - the process is dead and
173 * can't be modifying its own credentials. But shut RCU-lockdep up */
175 atomic_dec(&__task_cred(p)->user->processes);
180 write_lock_irq(&tasklist_lock);
181 ptrace_release_task(p);
185 * If we are the last non-leader member of the thread
186 * group, and the leader is zombie, then notify the
187 * group leader's parent process. (if it wants notification.)
190 leader = p->group_leader;
191 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
192 BUG_ON(task_detached(leader));
194 * If we were the last child thread and the leader has
195 * exited already, and the leader's parent ignores SIGCHLD,
196 * then we are the one who should release the leader.
198 zap_leader = do_notify_parent(leader, leader->exit_signal);
200 leader->exit_state = EXIT_DEAD;
203 write_unlock_irq(&tasklist_lock);
205 call_rcu(&p->rcu, delayed_put_task_struct);
208 if (unlikely(zap_leader))
213 * This checks not only the pgrp, but falls back on the pid if no
214 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
217 * The caller must hold rcu lock or the tasklist lock.
219 struct pid *session_of_pgrp(struct pid *pgrp)
221 struct task_struct *p;
222 struct pid *sid = NULL;
224 p = pid_task(pgrp, PIDTYPE_PGID);
226 p = pid_task(pgrp, PIDTYPE_PID);
228 sid = task_session(p);
234 * Determine if a process group is "orphaned", according to the POSIX
235 * definition in 2.2.2.52. Orphaned process groups are not to be affected
236 * by terminal-generated stop signals. Newly orphaned process groups are
237 * to receive a SIGHUP and a SIGCONT.
239 * "I ask you, have you ever known what it is to be an orphan?"
241 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
243 struct task_struct *p;
245 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
246 if ((p == ignored_task) ||
247 (p->exit_state && thread_group_empty(p)) ||
248 is_global_init(p->real_parent))
251 if (task_pgrp(p->real_parent) != pgrp &&
252 task_session(p->real_parent) == task_session(p))
254 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
259 int is_current_pgrp_orphaned(void)
263 read_lock(&tasklist_lock);
264 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
265 read_unlock(&tasklist_lock);
270 static int has_stopped_jobs(struct pid *pgrp)
273 struct task_struct *p;
275 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
276 if (!task_is_stopped(p))
280 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
285 * Check to see if any process groups have become orphaned as
286 * a result of our exiting, and if they have any stopped jobs,
287 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
290 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
292 struct pid *pgrp = task_pgrp(tsk);
293 struct task_struct *ignored_task = tsk;
296 /* exit: our father is in a different pgrp than
297 * we are and we were the only connection outside.
299 parent = tsk->real_parent;
301 /* reparent: our child is in a different pgrp than
302 * we are, and it was the only connection outside.
306 if (task_pgrp(parent) != pgrp &&
307 task_session(parent) == task_session(tsk) &&
308 will_become_orphaned_pgrp(pgrp, ignored_task) &&
309 has_stopped_jobs(pgrp)) {
310 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
311 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
316 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
318 * If a kernel thread is launched as a result of a system call, or if
319 * it ever exits, it should generally reparent itself to kthreadd so it
320 * isn't in the way of other processes and is correctly cleaned up on exit.
322 * The various task state such as scheduling policy and priority may have
323 * been inherited from a user process, so we reset them to sane values here.
325 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
327 static void reparent_to_kthreadd(void)
329 write_lock_irq(&tasklist_lock);
331 ptrace_unlink(current);
332 /* Reparent to init */
333 current->real_parent = current->parent = kthreadd_task;
334 list_move_tail(¤t->sibling, ¤t->real_parent->children);
336 /* Set the exit signal to SIGCHLD so we signal init on exit */
337 current->exit_signal = SIGCHLD;
339 if (task_nice(current) < 0)
340 set_user_nice(current, 0);
344 memcpy(current->signal->rlim, init_task.signal->rlim,
345 sizeof(current->signal->rlim));
347 atomic_inc(&init_cred.usage);
348 commit_creds(&init_cred);
349 write_unlock_irq(&tasklist_lock);
352 void __set_special_pids(struct pid *pid)
354 struct task_struct *curr = current->group_leader;
356 if (task_session(curr) != pid)
357 change_pid(curr, PIDTYPE_SID, pid);
359 if (task_pgrp(curr) != pid)
360 change_pid(curr, PIDTYPE_PGID, pid);
363 static void set_special_pids(struct pid *pid)
365 write_lock_irq(&tasklist_lock);
366 __set_special_pids(pid);
367 write_unlock_irq(&tasklist_lock);
371 * Let kernel threads use this to say that they allow a certain signal.
372 * Must not be used if kthread was cloned with CLONE_SIGHAND.
374 int allow_signal(int sig)
376 if (!valid_signal(sig) || sig < 1)
379 spin_lock_irq(¤t->sighand->siglock);
380 /* This is only needed for daemonize()'ed kthreads */
381 sigdelset(¤t->blocked, sig);
383 * Kernel threads handle their own signals. Let the signal code
384 * know it'll be handled, so that they don't get converted to
385 * SIGKILL or just silently dropped.
387 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
389 spin_unlock_irq(¤t->sighand->siglock);
393 EXPORT_SYMBOL(allow_signal);
395 int disallow_signal(int sig)
397 if (!valid_signal(sig) || sig < 1)
400 spin_lock_irq(¤t->sighand->siglock);
401 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
403 spin_unlock_irq(¤t->sighand->siglock);
407 EXPORT_SYMBOL(disallow_signal);
410 * Put all the gunge required to become a kernel thread without
411 * attached user resources in one place where it belongs.
414 void daemonize(const char *name, ...)
419 va_start(args, name);
420 vsnprintf(current->comm, sizeof(current->comm), name, args);
424 * If we were started as result of loading a module, close all of the
425 * user space pages. We don't need them, and if we didn't close them
426 * they would be locked into memory.
430 * We don't want to have TIF_FREEZE set if the system-wide hibernation
431 * or suspend transition begins right now.
433 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
435 if (current->nsproxy != &init_nsproxy) {
436 get_nsproxy(&init_nsproxy);
437 switch_task_namespaces(current, &init_nsproxy);
439 set_special_pids(&init_struct_pid);
440 proc_clear_tty(current);
442 /* Block and flush all signals */
443 sigfillset(&blocked);
444 sigprocmask(SIG_BLOCK, &blocked, NULL);
445 flush_signals(current);
447 /* Become as one with the init task */
449 daemonize_fs_struct();
451 current->files = init_task.files;
452 atomic_inc(¤t->files->count);
454 reparent_to_kthreadd();
457 EXPORT_SYMBOL(daemonize);
459 static void close_files(struct files_struct * files)
467 * It is safe to dereference the fd table without RCU or
468 * ->file_lock because this is the last reference to the
469 * files structure. But use RCU to shut RCU-lockdep up.
472 fdt = files_fdtable(files);
477 if (i >= fdt->max_fds)
479 set = fdt->open_fds->fds_bits[j++];
482 struct file * file = xchg(&fdt->fd[i], NULL);
484 filp_close(file, files);
494 struct files_struct *get_files_struct(struct task_struct *task)
496 struct files_struct *files;
501 atomic_inc(&files->count);
507 void put_files_struct(struct files_struct *files)
511 if (atomic_dec_and_test(&files->count)) {
514 * Free the fd and fdset arrays if we expanded them.
515 * If the fdtable was embedded, pass files for freeing
516 * at the end of the RCU grace period. Otherwise,
517 * you can free files immediately.
520 fdt = files_fdtable(files);
521 if (fdt != &files->fdtab)
522 kmem_cache_free(files_cachep, files);
528 void reset_files_struct(struct files_struct *files)
530 struct task_struct *tsk = current;
531 struct files_struct *old;
537 put_files_struct(old);
540 void exit_files(struct task_struct *tsk)
542 struct files_struct * files = tsk->files;
548 put_files_struct(files);
552 #ifdef CONFIG_MM_OWNER
554 * Task p is exiting and it owned mm, lets find a new owner for it
557 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
560 * If there are other users of the mm and the owner (us) is exiting
561 * we need to find a new owner to take on the responsibility.
563 if (atomic_read(&mm->mm_users) <= 1)
570 void mm_update_next_owner(struct mm_struct *mm)
572 struct task_struct *c, *g, *p = current;
575 if (!mm_need_new_owner(mm, p))
578 read_lock(&tasklist_lock);
580 * Search in the children
582 list_for_each_entry(c, &p->children, sibling) {
584 goto assign_new_owner;
588 * Search in the siblings
590 list_for_each_entry(c, &p->real_parent->children, sibling) {
592 goto assign_new_owner;
596 * Search through everything else. We should not get
599 do_each_thread(g, c) {
601 goto assign_new_owner;
602 } while_each_thread(g, c);
604 read_unlock(&tasklist_lock);
606 * We found no owner yet mm_users > 1: this implies that we are
607 * most likely racing with swapoff (try_to_unuse()) or /proc or
608 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
617 * The task_lock protects c->mm from changing.
618 * We always want mm->owner->mm == mm
622 * Delay read_unlock() till we have the task_lock()
623 * to ensure that c does not slip away underneath us
625 read_unlock(&tasklist_lock);
635 #endif /* CONFIG_MM_OWNER */
638 * Turn us into a lazy TLB process if we
641 static void exit_mm(struct task_struct * tsk)
643 struct mm_struct *mm = tsk->mm;
644 struct core_state *core_state;
650 * Serialize with any possible pending coredump.
651 * We must hold mmap_sem around checking core_state
652 * and clearing tsk->mm. The core-inducing thread
653 * will increment ->nr_threads for each thread in the
654 * group with ->mm != NULL.
656 down_read(&mm->mmap_sem);
657 core_state = mm->core_state;
659 struct core_thread self;
660 up_read(&mm->mmap_sem);
663 self.next = xchg(&core_state->dumper.next, &self);
665 * Implies mb(), the result of xchg() must be visible
666 * to core_state->dumper.
668 if (atomic_dec_and_test(&core_state->nr_threads))
669 complete(&core_state->startup);
672 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
673 if (!self.task) /* see coredump_finish() */
677 __set_task_state(tsk, TASK_RUNNING);
678 down_read(&mm->mmap_sem);
680 atomic_inc(&mm->mm_count);
681 BUG_ON(mm != tsk->active_mm);
682 /* more a memory barrier than a real lock */
685 up_read(&mm->mmap_sem);
686 enter_lazy_tlb(mm, current);
687 /* We don't want this task to be frozen prematurely */
688 clear_freeze_flag(tsk);
689 if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
690 atomic_dec(&mm->oom_disable_count);
692 mm_update_next_owner(mm);
697 * When we die, we re-parent all our children.
698 * Try to give them to another thread in our thread
699 * group, and if no such member exists, give it to
700 * the child reaper process (ie "init") in our pid
703 static struct task_struct *find_new_reaper(struct task_struct *father)
704 __releases(&tasklist_lock)
705 __acquires(&tasklist_lock)
707 struct pid_namespace *pid_ns = task_active_pid_ns(father);
708 struct task_struct *thread;
711 while_each_thread(father, thread) {
712 if (thread->flags & PF_EXITING)
714 if (unlikely(pid_ns->child_reaper == father))
715 pid_ns->child_reaper = thread;
719 if (unlikely(pid_ns->child_reaper == father)) {
720 write_unlock_irq(&tasklist_lock);
721 if (unlikely(pid_ns == &init_pid_ns))
722 panic("Attempted to kill init!");
724 zap_pid_ns_processes(pid_ns);
725 write_lock_irq(&tasklist_lock);
727 * We can not clear ->child_reaper or leave it alone.
728 * There may by stealth EXIT_DEAD tasks on ->children,
729 * forget_original_parent() must move them somewhere.
731 pid_ns->child_reaper = init_pid_ns.child_reaper;
734 return pid_ns->child_reaper;
738 * Any that need to be release_task'd are put on the @dead list.
740 static void reparent_leader(struct task_struct *father, struct task_struct *p,
741 struct list_head *dead)
743 list_move_tail(&p->sibling, &p->real_parent->children);
745 if (task_detached(p))
748 * If this is a threaded reparent there is no need to
749 * notify anyone anything has happened.
751 if (same_thread_group(p->real_parent, father))
754 /* We don't want people slaying init. */
755 p->exit_signal = SIGCHLD;
757 /* If it has exited notify the new parent about this child's death. */
759 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
760 if (do_notify_parent(p, p->exit_signal)) {
761 p->exit_state = EXIT_DEAD;
762 list_move_tail(&p->sibling, dead);
766 kill_orphaned_pgrp(p, father);
769 static void forget_original_parent(struct task_struct *father)
771 struct task_struct *p, *n, *reaper;
772 LIST_HEAD(dead_children);
774 write_lock_irq(&tasklist_lock);
776 * Note that exit_ptrace() and find_new_reaper() might
777 * drop tasklist_lock and reacquire it.
780 reaper = find_new_reaper(father);
782 list_for_each_entry_safe(p, n, &father->children, sibling) {
783 struct task_struct *t = p;
785 t->real_parent = reaper;
786 if (t->parent == father) {
788 t->parent = t->real_parent;
790 if (t->pdeath_signal)
791 group_send_sig_info(t->pdeath_signal,
793 } while_each_thread(p, t);
794 reparent_leader(father, p, &dead_children);
796 write_unlock_irq(&tasklist_lock);
798 BUG_ON(!list_empty(&father->children));
800 list_for_each_entry_safe(p, n, &dead_children, sibling) {
801 list_del_init(&p->sibling);
807 * Send signals to all our closest relatives so that they know
808 * to properly mourn us..
810 static void exit_notify(struct task_struct *tsk, int group_dead)
815 * This does two things:
817 * A. Make init inherit all the child processes
818 * B. Check to see if any process groups have become orphaned
819 * as a result of our exiting, and if they have any stopped
820 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
822 forget_original_parent(tsk);
823 exit_task_namespaces(tsk);
825 write_lock_irq(&tasklist_lock);
827 kill_orphaned_pgrp(tsk->group_leader, NULL);
829 /* Let father know we died
831 * Thread signals are configurable, but you aren't going to use
832 * that to send signals to arbitrary processes.
833 * That stops right now.
835 * If the parent exec id doesn't match the exec id we saved
836 * when we started then we know the parent has changed security
839 * If our self_exec id doesn't match our parent_exec_id then
840 * we have changed execution domain as these two values started
841 * the same after a fork.
843 if (thread_group_leader(tsk) && tsk->exit_signal != SIGCHLD &&
844 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
845 tsk->self_exec_id != tsk->parent_exec_id))
846 tsk->exit_signal = SIGCHLD;
848 if (unlikely(tsk->ptrace)) {
849 int sig = thread_group_leader(tsk) &&
850 thread_group_empty(tsk) &&
851 !ptrace_reparented(tsk) ?
852 tsk->exit_signal : SIGCHLD;
853 autoreap = do_notify_parent(tsk, sig);
854 } else if (thread_group_leader(tsk)) {
855 autoreap = thread_group_empty(tsk) &&
856 do_notify_parent(tsk, tsk->exit_signal);
861 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
863 /* mt-exec, de_thread() is waiting for group leader */
864 if (unlikely(tsk->signal->notify_count < 0))
865 wake_up_process(tsk->signal->group_exit_task);
866 write_unlock_irq(&tasklist_lock);
868 /* If the process is dead, release it - nobody will wait for it */
873 #ifdef CONFIG_DEBUG_STACK_USAGE
874 static void check_stack_usage(void)
876 static DEFINE_SPINLOCK(low_water_lock);
877 static int lowest_to_date = THREAD_SIZE;
880 free = stack_not_used(current);
882 if (free >= lowest_to_date)
885 spin_lock(&low_water_lock);
886 if (free < lowest_to_date) {
887 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
889 current->comm, free);
890 lowest_to_date = free;
892 spin_unlock(&low_water_lock);
895 static inline void check_stack_usage(void) {}
898 NORET_TYPE void do_exit(long code)
900 struct task_struct *tsk = current;
903 profile_task_exit(tsk);
905 WARN_ON(atomic_read(&tsk->fs_excl));
906 WARN_ON(blk_needs_flush_plug(tsk));
908 if (unlikely(in_interrupt()))
909 panic("Aiee, killing interrupt handler!");
910 if (unlikely(!tsk->pid))
911 panic("Attempted to kill the idle task!");
914 * If do_exit is called because this processes oopsed, it's possible
915 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
916 * continuing. Amongst other possible reasons, this is to prevent
917 * mm_release()->clear_child_tid() from writing to a user-controlled
922 ptrace_event(PTRACE_EVENT_EXIT, code);
924 validate_creds_for_do_exit(tsk);
927 * We're taking recursive faults here in do_exit. Safest is to just
928 * leave this task alone and wait for reboot.
930 if (unlikely(tsk->flags & PF_EXITING)) {
932 "Fixing recursive fault but reboot is needed!\n");
934 * We can do this unlocked here. The futex code uses
935 * this flag just to verify whether the pi state
936 * cleanup has been done or not. In the worst case it
937 * loops once more. We pretend that the cleanup was
938 * done as there is no way to return. Either the
939 * OWNER_DIED bit is set by now or we push the blocked
940 * task into the wait for ever nirwana as well.
942 tsk->flags |= PF_EXITPIDONE;
943 set_current_state(TASK_UNINTERRUPTIBLE);
949 exit_signals(tsk); /* sets PF_EXITING */
951 * tsk->flags are checked in the futex code to protect against
952 * an exiting task cleaning up the robust pi futexes.
955 raw_spin_unlock_wait(&tsk->pi_lock);
957 if (unlikely(in_atomic()))
958 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
959 current->comm, task_pid_nr(current),
962 acct_update_integrals(tsk);
963 /* sync mm's RSS info before statistics gathering */
965 sync_mm_rss(tsk, tsk->mm);
966 group_dead = atomic_dec_and_test(&tsk->signal->live);
968 hrtimer_cancel(&tsk->signal->real_timer);
969 exit_itimers(tsk->signal);
971 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
973 acct_collect(code, group_dead);
976 if (unlikely(tsk->audit_context))
979 tsk->exit_code = code;
980 taskstats_exit(tsk, group_dead);
986 trace_sched_process_exit(tsk);
995 * Flush inherited counters to the parent - before the parent
996 * gets woken up by child-exit notifications.
998 * because of cgroup mode, must be called before cgroup_exit()
1000 perf_event_exit_task(tsk);
1002 cgroup_exit(tsk, 1);
1005 disassociate_ctty(1);
1007 module_put(task_thread_info(tsk)->exec_domain->module);
1009 proc_exit_connector(tsk);
1012 * FIXME: do that only when needed, using sched_exit tracepoint
1014 ptrace_put_breakpoints(tsk);
1016 exit_notify(tsk, group_dead);
1019 mpol_put(tsk->mempolicy);
1020 tsk->mempolicy = NULL;
1024 if (unlikely(current->pi_state_cache))
1025 kfree(current->pi_state_cache);
1028 * Make sure we are holding no locks:
1030 debug_check_no_locks_held(tsk);
1032 * We can do this unlocked here. The futex code uses this flag
1033 * just to verify whether the pi state cleanup has been done
1034 * or not. In the worst case it loops once more.
1036 tsk->flags |= PF_EXITPIDONE;
1038 if (tsk->io_context)
1039 exit_io_context(tsk);
1041 if (tsk->splice_pipe)
1042 __free_pipe_info(tsk->splice_pipe);
1044 validate_creds_for_do_exit(tsk);
1048 /* causes final put_task_struct in finish_task_switch(). */
1049 tsk->state = TASK_DEAD;
1052 /* Avoid "noreturn function does return". */
1054 cpu_relax(); /* For when BUG is null */
1057 EXPORT_SYMBOL_GPL(do_exit);
1059 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1067 EXPORT_SYMBOL(complete_and_exit);
1069 SYSCALL_DEFINE1(exit, int, error_code)
1071 do_exit((error_code&0xff)<<8);
1075 * Take down every thread in the group. This is called by fatal signals
1076 * as well as by sys_exit_group (below).
1079 do_group_exit(int exit_code)
1081 struct signal_struct *sig = current->signal;
1083 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1085 if (signal_group_exit(sig))
1086 exit_code = sig->group_exit_code;
1087 else if (!thread_group_empty(current)) {
1088 struct sighand_struct *const sighand = current->sighand;
1089 spin_lock_irq(&sighand->siglock);
1090 if (signal_group_exit(sig))
1091 /* Another thread got here before we took the lock. */
1092 exit_code = sig->group_exit_code;
1094 sig->group_exit_code = exit_code;
1095 sig->flags = SIGNAL_GROUP_EXIT;
1096 zap_other_threads(current);
1098 spin_unlock_irq(&sighand->siglock);
1106 * this kills every thread in the thread group. Note that any externally
1107 * wait4()-ing process will get the correct exit code - even if this
1108 * thread is not the thread group leader.
1110 SYSCALL_DEFINE1(exit_group, int, error_code)
1112 do_group_exit((error_code & 0xff) << 8);
1118 enum pid_type wo_type;
1122 struct siginfo __user *wo_info;
1123 int __user *wo_stat;
1124 struct rusage __user *wo_rusage;
1126 wait_queue_t child_wait;
1131 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1133 if (type != PIDTYPE_PID)
1134 task = task->group_leader;
1135 return task->pids[type].pid;
1138 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1140 return wo->wo_type == PIDTYPE_MAX ||
1141 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1144 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1146 if (!eligible_pid(wo, p))
1148 /* Wait for all children (clone and not) if __WALL is set;
1149 * otherwise, wait for clone children *only* if __WCLONE is
1150 * set; otherwise, wait for non-clone children *only*. (Note:
1151 * A "clone" child here is one that reports to its parent
1152 * using a signal other than SIGCHLD.) */
1153 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1154 && !(wo->wo_flags & __WALL))
1160 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1161 pid_t pid, uid_t uid, int why, int status)
1163 struct siginfo __user *infop;
1164 int retval = wo->wo_rusage
1165 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1168 infop = wo->wo_info;
1171 retval = put_user(SIGCHLD, &infop->si_signo);
1173 retval = put_user(0, &infop->si_errno);
1175 retval = put_user((short)why, &infop->si_code);
1177 retval = put_user(pid, &infop->si_pid);
1179 retval = put_user(uid, &infop->si_uid);
1181 retval = put_user(status, &infop->si_status);
1189 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1190 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1191 * the lock and this task is uninteresting. If we return nonzero, we have
1192 * released the lock and the system call should return.
1194 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1196 unsigned long state;
1197 int retval, status, traced;
1198 pid_t pid = task_pid_vnr(p);
1199 uid_t uid = __task_cred(p)->uid;
1200 struct siginfo __user *infop;
1202 if (!likely(wo->wo_flags & WEXITED))
1205 if (unlikely(wo->wo_flags & WNOWAIT)) {
1206 int exit_code = p->exit_code;
1210 read_unlock(&tasklist_lock);
1211 if ((exit_code & 0x7f) == 0) {
1213 status = exit_code >> 8;
1215 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1216 status = exit_code & 0x7f;
1218 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1222 * Try to move the task's state to DEAD
1223 * only one thread is allowed to do this:
1225 state = xchg(&p->exit_state, EXIT_DEAD);
1226 if (state != EXIT_ZOMBIE) {
1227 BUG_ON(state != EXIT_DEAD);
1231 traced = ptrace_reparented(p);
1233 * It can be ptraced but not reparented, check
1234 * !task_detached() to filter out sub-threads.
1236 if (likely(!traced) && likely(!task_detached(p))) {
1237 struct signal_struct *psig;
1238 struct signal_struct *sig;
1239 unsigned long maxrss;
1240 cputime_t tgutime, tgstime;
1243 * The resource counters for the group leader are in its
1244 * own task_struct. Those for dead threads in the group
1245 * are in its signal_struct, as are those for the child
1246 * processes it has previously reaped. All these
1247 * accumulate in the parent's signal_struct c* fields.
1249 * We don't bother to take a lock here to protect these
1250 * p->signal fields, because they are only touched by
1251 * __exit_signal, which runs with tasklist_lock
1252 * write-locked anyway, and so is excluded here. We do
1253 * need to protect the access to parent->signal fields,
1254 * as other threads in the parent group can be right
1255 * here reaping other children at the same time.
1257 * We use thread_group_times() to get times for the thread
1258 * group, which consolidates times for all threads in the
1259 * group including the group leader.
1261 thread_group_times(p, &tgutime, &tgstime);
1262 spin_lock_irq(&p->real_parent->sighand->siglock);
1263 psig = p->real_parent->signal;
1266 cputime_add(psig->cutime,
1267 cputime_add(tgutime,
1270 cputime_add(psig->cstime,
1271 cputime_add(tgstime,
1274 cputime_add(psig->cgtime,
1275 cputime_add(p->gtime,
1276 cputime_add(sig->gtime,
1279 p->min_flt + sig->min_flt + sig->cmin_flt;
1281 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1283 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1285 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1287 task_io_get_inblock(p) +
1288 sig->inblock + sig->cinblock;
1290 task_io_get_oublock(p) +
1291 sig->oublock + sig->coublock;
1292 maxrss = max(sig->maxrss, sig->cmaxrss);
1293 if (psig->cmaxrss < maxrss)
1294 psig->cmaxrss = maxrss;
1295 task_io_accounting_add(&psig->ioac, &p->ioac);
1296 task_io_accounting_add(&psig->ioac, &sig->ioac);
1297 spin_unlock_irq(&p->real_parent->sighand->siglock);
1301 * Now we are sure this task is interesting, and no other
1302 * thread can reap it because we set its state to EXIT_DEAD.
1304 read_unlock(&tasklist_lock);
1306 retval = wo->wo_rusage
1307 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1308 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1309 ? p->signal->group_exit_code : p->exit_code;
1310 if (!retval && wo->wo_stat)
1311 retval = put_user(status, wo->wo_stat);
1313 infop = wo->wo_info;
1314 if (!retval && infop)
1315 retval = put_user(SIGCHLD, &infop->si_signo);
1316 if (!retval && infop)
1317 retval = put_user(0, &infop->si_errno);
1318 if (!retval && infop) {
1321 if ((status & 0x7f) == 0) {
1325 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1328 retval = put_user((short)why, &infop->si_code);
1330 retval = put_user(status, &infop->si_status);
1332 if (!retval && infop)
1333 retval = put_user(pid, &infop->si_pid);
1334 if (!retval && infop)
1335 retval = put_user(uid, &infop->si_uid);
1340 write_lock_irq(&tasklist_lock);
1341 /* We dropped tasklist, ptracer could die and untrace */
1344 * If this is not a sub-thread, notify the parent.
1345 * If parent wants a zombie, don't release it now.
1347 if (thread_group_leader(p) &&
1348 !do_notify_parent(p, p->exit_signal)) {
1349 p->exit_state = EXIT_ZOMBIE;
1352 write_unlock_irq(&tasklist_lock);
1360 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1363 if (task_is_stopped_or_traced(p) &&
1364 !(p->jobctl & JOBCTL_LISTENING))
1365 return &p->exit_code;
1367 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1368 return &p->signal->group_exit_code;
1374 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1376 * @ptrace: is the wait for ptrace
1377 * @p: task to wait for
1379 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1382 * read_lock(&tasklist_lock), which is released if return value is
1383 * non-zero. Also, grabs and releases @p->sighand->siglock.
1386 * 0 if wait condition didn't exist and search for other wait conditions
1387 * should continue. Non-zero return, -errno on failure and @p's pid on
1388 * success, implies that tasklist_lock is released and wait condition
1389 * search should terminate.
1391 static int wait_task_stopped(struct wait_opts *wo,
1392 int ptrace, struct task_struct *p)
1394 struct siginfo __user *infop;
1395 int retval, exit_code, *p_code, why;
1396 uid_t uid = 0; /* unneeded, required by compiler */
1400 * Traditionally we see ptrace'd stopped tasks regardless of options.
1402 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1405 if (!task_stopped_code(p, ptrace))
1409 spin_lock_irq(&p->sighand->siglock);
1411 p_code = task_stopped_code(p, ptrace);
1412 if (unlikely(!p_code))
1415 exit_code = *p_code;
1419 if (!unlikely(wo->wo_flags & WNOWAIT))
1424 spin_unlock_irq(&p->sighand->siglock);
1429 * Now we are pretty sure this task is interesting.
1430 * Make sure it doesn't get reaped out from under us while we
1431 * give up the lock and then examine it below. We don't want to
1432 * keep holding onto the tasklist_lock while we call getrusage and
1433 * possibly take page faults for user memory.
1436 pid = task_pid_vnr(p);
1437 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1438 read_unlock(&tasklist_lock);
1440 if (unlikely(wo->wo_flags & WNOWAIT))
1441 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1443 retval = wo->wo_rusage
1444 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1445 if (!retval && wo->wo_stat)
1446 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1448 infop = wo->wo_info;
1449 if (!retval && infop)
1450 retval = put_user(SIGCHLD, &infop->si_signo);
1451 if (!retval && infop)
1452 retval = put_user(0, &infop->si_errno);
1453 if (!retval && infop)
1454 retval = put_user((short)why, &infop->si_code);
1455 if (!retval && infop)
1456 retval = put_user(exit_code, &infop->si_status);
1457 if (!retval && infop)
1458 retval = put_user(pid, &infop->si_pid);
1459 if (!retval && infop)
1460 retval = put_user(uid, &infop->si_uid);
1470 * Handle do_wait work for one task in a live, non-stopped state.
1471 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1472 * the lock and this task is uninteresting. If we return nonzero, we have
1473 * released the lock and the system call should return.
1475 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1481 if (!unlikely(wo->wo_flags & WCONTINUED))
1484 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1487 spin_lock_irq(&p->sighand->siglock);
1488 /* Re-check with the lock held. */
1489 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1490 spin_unlock_irq(&p->sighand->siglock);
1493 if (!unlikely(wo->wo_flags & WNOWAIT))
1494 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1496 spin_unlock_irq(&p->sighand->siglock);
1498 pid = task_pid_vnr(p);
1500 read_unlock(&tasklist_lock);
1503 retval = wo->wo_rusage
1504 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1506 if (!retval && wo->wo_stat)
1507 retval = put_user(0xffff, wo->wo_stat);
1511 retval = wait_noreap_copyout(wo, p, pid, uid,
1512 CLD_CONTINUED, SIGCONT);
1513 BUG_ON(retval == 0);
1520 * Consider @p for a wait by @parent.
1522 * -ECHILD should be in ->notask_error before the first call.
1523 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1524 * Returns zero if the search for a child should continue;
1525 * then ->notask_error is 0 if @p is an eligible child,
1526 * or another error from security_task_wait(), or still -ECHILD.
1528 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1529 struct task_struct *p)
1531 int ret = eligible_child(wo, p);
1535 ret = security_task_wait(p);
1536 if (unlikely(ret < 0)) {
1538 * If we have not yet seen any eligible child,
1539 * then let this error code replace -ECHILD.
1540 * A permission error will give the user a clue
1541 * to look for security policy problems, rather
1542 * than for mysterious wait bugs.
1544 if (wo->notask_error)
1545 wo->notask_error = ret;
1549 /* dead body doesn't have much to contribute */
1550 if (p->exit_state == EXIT_DEAD)
1554 if (p->exit_state == EXIT_ZOMBIE) {
1556 * A zombie ptracee is only visible to its ptracer.
1557 * Notification and reaping will be cascaded to the real
1558 * parent when the ptracer detaches.
1560 if (likely(!ptrace) && unlikely(p->ptrace)) {
1561 /* it will become visible, clear notask_error */
1562 wo->notask_error = 0;
1566 /* we don't reap group leaders with subthreads */
1567 if (!delay_group_leader(p))
1568 return wait_task_zombie(wo, p);
1571 * Allow access to stopped/continued state via zombie by
1572 * falling through. Clearing of notask_error is complex.
1576 * If WEXITED is set, notask_error should naturally be
1577 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1578 * so, if there are live subthreads, there are events to
1579 * wait for. If all subthreads are dead, it's still safe
1580 * to clear - this function will be called again in finite
1581 * amount time once all the subthreads are released and
1582 * will then return without clearing.
1586 * Stopped state is per-task and thus can't change once the
1587 * target task dies. Only continued and exited can happen.
1588 * Clear notask_error if WCONTINUED | WEXITED.
1590 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1591 wo->notask_error = 0;
1594 * If @p is ptraced by a task in its real parent's group,
1595 * hide group stop/continued state when looking at @p as
1596 * the real parent; otherwise, a single stop can be
1597 * reported twice as group and ptrace stops.
1599 * If a ptracer wants to distinguish the two events for its
1600 * own children, it should create a separate process which
1601 * takes the role of real parent.
1603 if (likely(!ptrace) && p->ptrace &&
1604 same_thread_group(p->parent, p->real_parent))
1608 * @p is alive and it's gonna stop, continue or exit, so
1609 * there always is something to wait for.
1611 wo->notask_error = 0;
1615 * Wait for stopped. Depending on @ptrace, different stopped state
1616 * is used and the two don't interact with each other.
1618 ret = wait_task_stopped(wo, ptrace, p);
1623 * Wait for continued. There's only one continued state and the
1624 * ptracer can consume it which can confuse the real parent. Don't
1625 * use WCONTINUED from ptracer. You don't need or want it.
1627 return wait_task_continued(wo, p);
1631 * Do the work of do_wait() for one thread in the group, @tsk.
1633 * -ECHILD should be in ->notask_error before the first call.
1634 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1635 * Returns zero if the search for a child should continue; then
1636 * ->notask_error is 0 if there were any eligible children,
1637 * or another error from security_task_wait(), or still -ECHILD.
1639 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1641 struct task_struct *p;
1643 list_for_each_entry(p, &tsk->children, sibling) {
1644 int ret = wait_consider_task(wo, 0, p);
1652 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1654 struct task_struct *p;
1656 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1657 int ret = wait_consider_task(wo, 1, p);
1665 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1666 int sync, void *key)
1668 struct wait_opts *wo = container_of(wait, struct wait_opts,
1670 struct task_struct *p = key;
1672 if (!eligible_pid(wo, p))
1675 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1678 return default_wake_function(wait, mode, sync, key);
1681 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1683 __wake_up_sync_key(&parent->signal->wait_chldexit,
1684 TASK_INTERRUPTIBLE, 1, p);
1687 static long do_wait(struct wait_opts *wo)
1689 struct task_struct *tsk;
1692 trace_sched_process_wait(wo->wo_pid);
1694 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1695 wo->child_wait.private = current;
1696 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1699 * If there is nothing that can match our critiera just get out.
1700 * We will clear ->notask_error to zero if we see any child that
1701 * might later match our criteria, even if we are not able to reap
1704 wo->notask_error = -ECHILD;
1705 if ((wo->wo_type < PIDTYPE_MAX) &&
1706 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1709 set_current_state(TASK_INTERRUPTIBLE);
1710 read_lock(&tasklist_lock);
1713 retval = do_wait_thread(wo, tsk);
1717 retval = ptrace_do_wait(wo, tsk);
1721 if (wo->wo_flags & __WNOTHREAD)
1723 } while_each_thread(current, tsk);
1724 read_unlock(&tasklist_lock);
1727 retval = wo->notask_error;
1728 if (!retval && !(wo->wo_flags & WNOHANG)) {
1729 retval = -ERESTARTSYS;
1730 if (!signal_pending(current)) {
1736 __set_current_state(TASK_RUNNING);
1737 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1741 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1742 infop, int, options, struct rusage __user *, ru)
1744 struct wait_opts wo;
1745 struct pid *pid = NULL;
1749 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1751 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1764 type = PIDTYPE_PGID;
1772 if (type < PIDTYPE_MAX)
1773 pid = find_get_pid(upid);
1777 wo.wo_flags = options;
1787 * For a WNOHANG return, clear out all the fields
1788 * we would set so the user can easily tell the
1792 ret = put_user(0, &infop->si_signo);
1794 ret = put_user(0, &infop->si_errno);
1796 ret = put_user(0, &infop->si_code);
1798 ret = put_user(0, &infop->si_pid);
1800 ret = put_user(0, &infop->si_uid);
1802 ret = put_user(0, &infop->si_status);
1807 /* avoid REGPARM breakage on x86: */
1808 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1812 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1813 int, options, struct rusage __user *, ru)
1815 struct wait_opts wo;
1816 struct pid *pid = NULL;
1820 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1821 __WNOTHREAD|__WCLONE|__WALL))
1826 else if (upid < 0) {
1827 type = PIDTYPE_PGID;
1828 pid = find_get_pid(-upid);
1829 } else if (upid == 0) {
1830 type = PIDTYPE_PGID;
1831 pid = get_task_pid(current, PIDTYPE_PGID);
1832 } else /* upid > 0 */ {
1834 pid = find_get_pid(upid);
1839 wo.wo_flags = options | WEXITED;
1841 wo.wo_stat = stat_addr;
1846 /* avoid REGPARM breakage on x86: */
1847 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1851 #ifdef __ARCH_WANT_SYS_WAITPID
1854 * sys_waitpid() remains for compatibility. waitpid() should be
1855 * implemented by calling sys_wait4() from libc.a.
1857 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1859 return sys_wait4(pid, stat_addr, options, NULL);