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>
53 #include <asm/uaccess.h>
54 #include <asm/unistd.h>
55 #include <asm/pgtable.h>
56 #include <asm/mmu_context.h>
58 static void exit_mm(struct task_struct * tsk);
60 static void __unhash_process(struct task_struct *p)
63 detach_pid(p, PIDTYPE_PID);
64 if (thread_group_leader(p)) {
65 detach_pid(p, PIDTYPE_PGID);
66 detach_pid(p, PIDTYPE_SID);
68 list_del_rcu(&p->tasks);
69 __get_cpu_var(process_counts)--;
71 list_del_rcu(&p->thread_group);
72 list_del_init(&p->sibling);
76 * This function expects the tasklist_lock write-locked.
78 static void __exit_signal(struct task_struct *tsk)
80 struct signal_struct *sig = tsk->signal;
81 struct sighand_struct *sighand;
84 BUG_ON(!atomic_read(&sig->count));
86 sighand = rcu_dereference(tsk->sighand);
87 spin_lock(&sighand->siglock);
89 posix_cpu_timers_exit(tsk);
90 if (atomic_dec_and_test(&sig->count))
91 posix_cpu_timers_exit_group(tsk);
94 * This can only happen if the caller is de_thread().
95 * FIXME: this is the temporary hack, we should teach
96 * posix-cpu-timers to handle this case correctly.
98 if (unlikely(has_group_leader_pid(tsk)))
99 posix_cpu_timers_exit_group(tsk);
102 * If there is any task waiting for the group exit
105 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
106 wake_up_process(sig->group_exit_task);
108 if (tsk == sig->curr_target)
109 sig->curr_target = next_thread(tsk);
111 * Accumulate here the counters for all threads but the
112 * group leader as they die, so they can be added into
113 * the process-wide totals when those are taken.
114 * The group leader stays around as a zombie as long
115 * as there are other threads. When it gets reaped,
116 * the exit.c code will add its counts into these totals.
117 * We won't ever get here for the group leader, since it
118 * will have been the last reference on the signal_struct.
120 sig->utime = cputime_add(sig->utime, tsk->utime);
121 sig->stime = cputime_add(sig->stime, tsk->stime);
122 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
123 sig->min_flt += tsk->min_flt;
124 sig->maj_flt += tsk->maj_flt;
125 sig->nvcsw += tsk->nvcsw;
126 sig->nivcsw += tsk->nivcsw;
127 sig->inblock += task_io_get_inblock(tsk);
128 sig->oublock += task_io_get_oublock(tsk);
129 task_io_accounting_add(&sig->ioac, &tsk->ioac);
130 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
131 sig = NULL; /* Marker for below. */
134 __unhash_process(tsk);
137 * Do this under ->siglock, we can race with another thread
138 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
140 flush_sigqueue(&tsk->pending);
144 spin_unlock(&sighand->siglock);
146 __cleanup_sighand(sighand);
147 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
149 flush_sigqueue(&sig->shared_pending);
150 taskstats_tgid_free(sig);
152 * Make sure ->signal can't go away under rq->lock,
153 * see account_group_exec_runtime().
155 task_rq_unlock_wait(tsk);
156 __cleanup_signal(sig);
160 static void delayed_put_task_struct(struct rcu_head *rhp)
162 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
164 #ifdef CONFIG_PERF_EVENTS
165 WARN_ON_ONCE(tsk->perf_event_ctxp);
167 trace_sched_process_free(tsk);
168 put_task_struct(tsk);
172 void release_task(struct task_struct * p)
174 struct task_struct *leader;
177 tracehook_prepare_release_task(p);
178 /* don't need to get the RCU readlock here - the process is dead and
179 * can't be modifying its own credentials */
180 atomic_dec(&__task_cred(p)->user->processes);
184 write_lock_irq(&tasklist_lock);
185 tracehook_finish_release_task(p);
189 * If we are the last non-leader member of the thread
190 * group, and the leader is zombie, then notify the
191 * group leader's parent process. (if it wants notification.)
194 leader = p->group_leader;
195 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
196 BUG_ON(task_detached(leader));
197 do_notify_parent(leader, leader->exit_signal);
199 * If we were the last child thread and the leader has
200 * exited already, and the leader's parent ignores SIGCHLD,
201 * then we are the one who should release the leader.
203 * do_notify_parent() will have marked it self-reaping in
206 zap_leader = task_detached(leader);
209 * This maintains the invariant that release_task()
210 * only runs on a task in EXIT_DEAD, just for sanity.
213 leader->exit_state = EXIT_DEAD;
216 write_unlock_irq(&tasklist_lock);
218 call_rcu(&p->rcu, delayed_put_task_struct);
221 if (unlikely(zap_leader))
226 * This checks not only the pgrp, but falls back on the pid if no
227 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
230 * The caller must hold rcu lock or the tasklist lock.
232 struct pid *session_of_pgrp(struct pid *pgrp)
234 struct task_struct *p;
235 struct pid *sid = NULL;
237 p = pid_task(pgrp, PIDTYPE_PGID);
239 p = pid_task(pgrp, PIDTYPE_PID);
241 sid = task_session(p);
247 * Determine if a process group is "orphaned", according to the POSIX
248 * definition in 2.2.2.52. Orphaned process groups are not to be affected
249 * by terminal-generated stop signals. Newly orphaned process groups are
250 * to receive a SIGHUP and a SIGCONT.
252 * "I ask you, have you ever known what it is to be an orphan?"
254 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
256 struct task_struct *p;
258 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
259 if ((p == ignored_task) ||
260 (p->exit_state && thread_group_empty(p)) ||
261 is_global_init(p->real_parent))
264 if (task_pgrp(p->real_parent) != pgrp &&
265 task_session(p->real_parent) == task_session(p))
267 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
272 int is_current_pgrp_orphaned(void)
276 read_lock(&tasklist_lock);
277 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
278 read_unlock(&tasklist_lock);
283 static int has_stopped_jobs(struct pid *pgrp)
286 struct task_struct *p;
288 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
289 if (!task_is_stopped(p))
293 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
298 * Check to see if any process groups have become orphaned as
299 * a result of our exiting, and if they have any stopped jobs,
300 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
303 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
305 struct pid *pgrp = task_pgrp(tsk);
306 struct task_struct *ignored_task = tsk;
309 /* exit: our father is in a different pgrp than
310 * we are and we were the only connection outside.
312 parent = tsk->real_parent;
314 /* reparent: our child is in a different pgrp than
315 * we are, and it was the only connection outside.
319 if (task_pgrp(parent) != pgrp &&
320 task_session(parent) == task_session(tsk) &&
321 will_become_orphaned_pgrp(pgrp, ignored_task) &&
322 has_stopped_jobs(pgrp)) {
323 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
324 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
329 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
331 * If a kernel thread is launched as a result of a system call, or if
332 * it ever exits, it should generally reparent itself to kthreadd so it
333 * isn't in the way of other processes and is correctly cleaned up on exit.
335 * The various task state such as scheduling policy and priority may have
336 * been inherited from a user process, so we reset them to sane values here.
338 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
340 static void reparent_to_kthreadd(void)
342 write_lock_irq(&tasklist_lock);
344 ptrace_unlink(current);
345 /* Reparent to init */
346 current->real_parent = current->parent = kthreadd_task;
347 list_move_tail(¤t->sibling, ¤t->real_parent->children);
349 /* Set the exit signal to SIGCHLD so we signal init on exit */
350 current->exit_signal = SIGCHLD;
352 if (task_nice(current) < 0)
353 set_user_nice(current, 0);
357 memcpy(current->signal->rlim, init_task.signal->rlim,
358 sizeof(current->signal->rlim));
360 atomic_inc(&init_cred.usage);
361 commit_creds(&init_cred);
362 write_unlock_irq(&tasklist_lock);
365 void __set_special_pids(struct pid *pid)
367 struct task_struct *curr = current->group_leader;
369 if (task_session(curr) != pid)
370 change_pid(curr, PIDTYPE_SID, pid);
372 if (task_pgrp(curr) != pid)
373 change_pid(curr, PIDTYPE_PGID, pid);
376 static void set_special_pids(struct pid *pid)
378 write_lock_irq(&tasklist_lock);
379 __set_special_pids(pid);
380 write_unlock_irq(&tasklist_lock);
384 * Let kernel threads use this to say that they allow a certain signal.
385 * Must not be used if kthread was cloned with CLONE_SIGHAND.
387 int allow_signal(int sig)
389 if (!valid_signal(sig) || sig < 1)
392 spin_lock_irq(¤t->sighand->siglock);
393 /* This is only needed for daemonize()'ed kthreads */
394 sigdelset(¤t->blocked, sig);
396 * Kernel threads handle their own signals. Let the signal code
397 * know it'll be handled, so that they don't get converted to
398 * SIGKILL or just silently dropped.
400 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
402 spin_unlock_irq(¤t->sighand->siglock);
406 EXPORT_SYMBOL(allow_signal);
408 int disallow_signal(int sig)
410 if (!valid_signal(sig) || sig < 1)
413 spin_lock_irq(¤t->sighand->siglock);
414 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
416 spin_unlock_irq(¤t->sighand->siglock);
420 EXPORT_SYMBOL(disallow_signal);
423 * Put all the gunge required to become a kernel thread without
424 * attached user resources in one place where it belongs.
427 void daemonize(const char *name, ...)
432 va_start(args, name);
433 vsnprintf(current->comm, sizeof(current->comm), name, args);
437 * If we were started as result of loading a module, close all of the
438 * user space pages. We don't need them, and if we didn't close them
439 * they would be locked into memory.
443 * We don't want to have TIF_FREEZE set if the system-wide hibernation
444 * or suspend transition begins right now.
446 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
448 if (current->nsproxy != &init_nsproxy) {
449 get_nsproxy(&init_nsproxy);
450 switch_task_namespaces(current, &init_nsproxy);
452 set_special_pids(&init_struct_pid);
453 proc_clear_tty(current);
455 /* Block and flush all signals */
456 sigfillset(&blocked);
457 sigprocmask(SIG_BLOCK, &blocked, NULL);
458 flush_signals(current);
460 /* Become as one with the init task */
462 daemonize_fs_struct();
464 current->files = init_task.files;
465 atomic_inc(¤t->files->count);
467 reparent_to_kthreadd();
470 EXPORT_SYMBOL(daemonize);
472 static void close_files(struct files_struct * files)
480 * It is safe to dereference the fd table without RCU or
481 * ->file_lock because this is the last reference to the
484 fdt = files_fdtable(files);
488 if (i >= fdt->max_fds)
490 set = fdt->open_fds->fds_bits[j++];
493 struct file * file = xchg(&fdt->fd[i], NULL);
495 filp_close(file, files);
505 struct files_struct *get_files_struct(struct task_struct *task)
507 struct files_struct *files;
512 atomic_inc(&files->count);
518 void put_files_struct(struct files_struct *files)
522 if (atomic_dec_and_test(&files->count)) {
525 * Free the fd and fdset arrays if we expanded them.
526 * If the fdtable was embedded, pass files for freeing
527 * at the end of the RCU grace period. Otherwise,
528 * you can free files immediately.
530 fdt = files_fdtable(files);
531 if (fdt != &files->fdtab)
532 kmem_cache_free(files_cachep, files);
537 void reset_files_struct(struct files_struct *files)
539 struct task_struct *tsk = current;
540 struct files_struct *old;
546 put_files_struct(old);
549 void exit_files(struct task_struct *tsk)
551 struct files_struct * files = tsk->files;
557 put_files_struct(files);
561 #ifdef CONFIG_MM_OWNER
563 * Task p is exiting and it owned mm, lets find a new owner for it
566 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
569 * If there are other users of the mm and the owner (us) is exiting
570 * we need to find a new owner to take on the responsibility.
572 if (atomic_read(&mm->mm_users) <= 1)
579 void mm_update_next_owner(struct mm_struct *mm)
581 struct task_struct *c, *g, *p = current;
584 if (!mm_need_new_owner(mm, p))
587 read_lock(&tasklist_lock);
589 * Search in the children
591 list_for_each_entry(c, &p->children, sibling) {
593 goto assign_new_owner;
597 * Search in the siblings
599 list_for_each_entry(c, &p->real_parent->children, sibling) {
601 goto assign_new_owner;
605 * Search through everything else. We should not get
608 do_each_thread(g, c) {
610 goto assign_new_owner;
611 } while_each_thread(g, c);
613 read_unlock(&tasklist_lock);
615 * We found no owner yet mm_users > 1: this implies that we are
616 * most likely racing with swapoff (try_to_unuse()) or /proc or
617 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
626 * The task_lock protects c->mm from changing.
627 * We always want mm->owner->mm == mm
631 * Delay read_unlock() till we have the task_lock()
632 * to ensure that c does not slip away underneath us
634 read_unlock(&tasklist_lock);
644 #endif /* CONFIG_MM_OWNER */
647 * Turn us into a lazy TLB process if we
650 static void exit_mm(struct task_struct * tsk)
652 struct mm_struct *mm = tsk->mm;
653 struct core_state *core_state;
659 * Serialize with any possible pending coredump.
660 * We must hold mmap_sem around checking core_state
661 * and clearing tsk->mm. The core-inducing thread
662 * will increment ->nr_threads for each thread in the
663 * group with ->mm != NULL.
665 down_read(&mm->mmap_sem);
666 core_state = mm->core_state;
668 struct core_thread self;
669 up_read(&mm->mmap_sem);
672 self.next = xchg(&core_state->dumper.next, &self);
674 * Implies mb(), the result of xchg() must be visible
675 * to core_state->dumper.
677 if (atomic_dec_and_test(&core_state->nr_threads))
678 complete(&core_state->startup);
681 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
682 if (!self.task) /* see coredump_finish() */
686 __set_task_state(tsk, TASK_RUNNING);
687 down_read(&mm->mmap_sem);
689 atomic_inc(&mm->mm_count);
690 BUG_ON(mm != tsk->active_mm);
691 /* more a memory barrier than a real lock */
694 up_read(&mm->mmap_sem);
695 enter_lazy_tlb(mm, current);
696 /* We don't want this task to be frozen prematurely */
697 clear_freeze_flag(tsk);
699 mm_update_next_owner(mm);
704 * When we die, we re-parent all our children.
705 * Try to give them to another thread in our thread
706 * group, and if no such member exists, give it to
707 * the child reaper process (ie "init") in our pid
710 static struct task_struct *find_new_reaper(struct task_struct *father)
712 struct pid_namespace *pid_ns = task_active_pid_ns(father);
713 struct task_struct *thread;
716 while_each_thread(father, thread) {
717 if (thread->flags & PF_EXITING)
719 if (unlikely(pid_ns->child_reaper == father))
720 pid_ns->child_reaper = thread;
724 if (unlikely(pid_ns->child_reaper == father)) {
725 write_unlock_irq(&tasklist_lock);
726 if (unlikely(pid_ns == &init_pid_ns))
727 panic("Attempted to kill init!");
729 zap_pid_ns_processes(pid_ns);
730 write_lock_irq(&tasklist_lock);
732 * We can not clear ->child_reaper or leave it alone.
733 * There may by stealth EXIT_DEAD tasks on ->children,
734 * forget_original_parent() must move them somewhere.
736 pid_ns->child_reaper = init_pid_ns.child_reaper;
739 return pid_ns->child_reaper;
743 * Any that need to be release_task'd are put on the @dead list.
745 static void reparent_thread(struct task_struct *father, struct task_struct *p,
746 struct list_head *dead)
748 if (p->pdeath_signal)
749 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
751 list_move_tail(&p->sibling, &p->real_parent->children);
753 if (task_detached(p))
756 * If this is a threaded reparent there is no need to
757 * notify anyone anything has happened.
759 if (same_thread_group(p->real_parent, father))
762 /* We don't want people slaying init. */
763 p->exit_signal = SIGCHLD;
765 /* If it has exited notify the new parent about this child's death. */
766 if (!task_ptrace(p) &&
767 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
768 do_notify_parent(p, p->exit_signal);
769 if (task_detached(p)) {
770 p->exit_state = EXIT_DEAD;
771 list_move_tail(&p->sibling, dead);
775 kill_orphaned_pgrp(p, father);
778 static void forget_original_parent(struct task_struct *father)
780 struct task_struct *p, *n, *reaper;
781 LIST_HEAD(dead_children);
785 write_lock_irq(&tasklist_lock);
786 reaper = find_new_reaper(father);
788 list_for_each_entry_safe(p, n, &father->children, sibling) {
789 p->real_parent = reaper;
790 if (p->parent == father) {
791 BUG_ON(task_ptrace(p));
792 p->parent = p->real_parent;
794 reparent_thread(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)
816 * This does two things:
818 * A. Make init inherit all the child processes
819 * B. Check to see if any process groups have become orphaned
820 * as a result of our exiting, and if they have any stopped
821 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
823 forget_original_parent(tsk);
824 exit_task_namespaces(tsk);
826 write_lock_irq(&tasklist_lock);
828 kill_orphaned_pgrp(tsk->group_leader, NULL);
830 /* Let father know we died
832 * Thread signals are configurable, but you aren't going to use
833 * that to send signals to arbitary processes.
834 * That stops right now.
836 * If the parent exec id doesn't match the exec id we saved
837 * when we started then we know the parent has changed security
840 * If our self_exec id doesn't match our parent_exec_id then
841 * we have changed execution domain as these two values started
842 * the same after a fork.
844 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
845 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
846 tsk->self_exec_id != tsk->parent_exec_id))
847 tsk->exit_signal = SIGCHLD;
849 signal = tracehook_notify_death(tsk, &cookie, group_dead);
851 signal = do_notify_parent(tsk, signal);
853 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
855 /* mt-exec, de_thread() is waiting for us */
856 if (thread_group_leader(tsk) &&
857 tsk->signal->group_exit_task &&
858 tsk->signal->notify_count < 0)
859 wake_up_process(tsk->signal->group_exit_task);
861 write_unlock_irq(&tasklist_lock);
863 tracehook_report_death(tsk, signal, cookie, group_dead);
865 /* If the process is dead, release it - nobody will wait for it */
866 if (signal == DEATH_REAP)
870 #ifdef CONFIG_DEBUG_STACK_USAGE
871 static void check_stack_usage(void)
873 static DEFINE_SPINLOCK(low_water_lock);
874 static int lowest_to_date = THREAD_SIZE;
877 free = stack_not_used(current);
879 if (free >= lowest_to_date)
882 spin_lock(&low_water_lock);
883 if (free < lowest_to_date) {
884 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
886 current->comm, free);
887 lowest_to_date = free;
889 spin_unlock(&low_water_lock);
892 static inline void check_stack_usage(void) {}
895 NORET_TYPE void do_exit(long code)
897 struct task_struct *tsk = current;
900 profile_task_exit(tsk);
902 WARN_ON(atomic_read(&tsk->fs_excl));
904 if (unlikely(in_interrupt()))
905 panic("Aiee, killing interrupt handler!");
906 if (unlikely(!tsk->pid))
907 panic("Attempted to kill the idle task!");
910 * If do_exit is called because this processes oopsed, it's possible
911 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
912 * continuing. Amongst other possible reasons, this is to prevent
913 * mm_release()->clear_child_tid() from writing to a user-controlled
918 tracehook_report_exit(&code);
920 validate_creds_for_do_exit(tsk);
923 * We're taking recursive faults here in do_exit. Safest is to just
924 * leave this task alone and wait for reboot.
926 if (unlikely(tsk->flags & PF_EXITING)) {
928 "Fixing recursive fault but reboot is needed!\n");
930 * We can do this unlocked here. The futex code uses
931 * this flag just to verify whether the pi state
932 * cleanup has been done or not. In the worst case it
933 * loops once more. We pretend that the cleanup was
934 * done as there is no way to return. Either the
935 * OWNER_DIED bit is set by now or we push the blocked
936 * task into the wait for ever nirwana as well.
938 tsk->flags |= PF_EXITPIDONE;
939 set_current_state(TASK_UNINTERRUPTIBLE);
945 exit_signals(tsk); /* sets PF_EXITING */
947 * tsk->flags are checked in the futex code to protect against
948 * an exiting task cleaning up the robust pi futexes.
951 spin_unlock_wait(&tsk->pi_lock);
953 if (unlikely(in_atomic()))
954 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
955 current->comm, task_pid_nr(current),
958 acct_update_integrals(tsk);
960 group_dead = atomic_dec_and_test(&tsk->signal->live);
962 hrtimer_cancel(&tsk->signal->real_timer);
963 exit_itimers(tsk->signal);
965 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
967 acct_collect(code, group_dead);
970 if (unlikely(tsk->audit_context))
973 tsk->exit_code = code;
974 taskstats_exit(tsk, group_dead);
980 trace_sched_process_exit(tsk);
989 if (group_dead && tsk->signal->leader)
990 disassociate_ctty(1);
992 module_put(task_thread_info(tsk)->exec_domain->module);
994 proc_exit_connector(tsk);
997 * Flush inherited counters to the parent - before the parent
998 * gets woken up by child-exit notifications.
1000 perf_event_exit_task(tsk);
1002 exit_notify(tsk, group_dead);
1004 mpol_put(tsk->mempolicy);
1005 tsk->mempolicy = NULL;
1008 if (unlikely(current->pi_state_cache))
1009 kfree(current->pi_state_cache);
1012 * Make sure we are holding no locks:
1014 debug_check_no_locks_held(tsk);
1016 * We can do this unlocked here. The futex code uses this flag
1017 * just to verify whether the pi state cleanup has been done
1018 * or not. In the worst case it loops once more.
1020 tsk->flags |= PF_EXITPIDONE;
1022 if (tsk->io_context)
1025 if (tsk->splice_pipe)
1026 __free_pipe_info(tsk->splice_pipe);
1028 validate_creds_for_do_exit(tsk);
1032 /* causes final put_task_struct in finish_task_switch(). */
1033 tsk->state = TASK_DEAD;
1036 /* Avoid "noreturn function does return". */
1038 cpu_relax(); /* For when BUG is null */
1041 EXPORT_SYMBOL_GPL(do_exit);
1043 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1051 EXPORT_SYMBOL(complete_and_exit);
1053 SYSCALL_DEFINE1(exit, int, error_code)
1055 do_exit((error_code&0xff)<<8);
1059 * Take down every thread in the group. This is called by fatal signals
1060 * as well as by sys_exit_group (below).
1063 do_group_exit(int exit_code)
1065 struct signal_struct *sig = current->signal;
1067 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1069 if (signal_group_exit(sig))
1070 exit_code = sig->group_exit_code;
1071 else if (!thread_group_empty(current)) {
1072 struct sighand_struct *const sighand = current->sighand;
1073 spin_lock_irq(&sighand->siglock);
1074 if (signal_group_exit(sig))
1075 /* Another thread got here before we took the lock. */
1076 exit_code = sig->group_exit_code;
1078 sig->group_exit_code = exit_code;
1079 sig->flags = SIGNAL_GROUP_EXIT;
1080 zap_other_threads(current);
1082 spin_unlock_irq(&sighand->siglock);
1090 * this kills every thread in the thread group. Note that any externally
1091 * wait4()-ing process will get the correct exit code - even if this
1092 * thread is not the thread group leader.
1094 SYSCALL_DEFINE1(exit_group, int, error_code)
1096 do_group_exit((error_code & 0xff) << 8);
1102 enum pid_type wo_type;
1106 struct siginfo __user *wo_info;
1107 int __user *wo_stat;
1108 struct rusage __user *wo_rusage;
1110 wait_queue_t child_wait;
1115 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1117 if (type != PIDTYPE_PID)
1118 task = task->group_leader;
1119 return task->pids[type].pid;
1122 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1124 return wo->wo_type == PIDTYPE_MAX ||
1125 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1128 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1130 if (!eligible_pid(wo, p))
1132 /* Wait for all children (clone and not) if __WALL is set;
1133 * otherwise, wait for clone children *only* if __WCLONE is
1134 * set; otherwise, wait for non-clone children *only*. (Note:
1135 * A "clone" child here is one that reports to its parent
1136 * using a signal other than SIGCHLD.) */
1137 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1138 && !(wo->wo_flags & __WALL))
1144 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1145 pid_t pid, uid_t uid, int why, int status)
1147 struct siginfo __user *infop;
1148 int retval = wo->wo_rusage
1149 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1152 infop = wo->wo_info;
1155 retval = put_user(SIGCHLD, &infop->si_signo);
1157 retval = put_user(0, &infop->si_errno);
1159 retval = put_user((short)why, &infop->si_code);
1161 retval = put_user(pid, &infop->si_pid);
1163 retval = put_user(uid, &infop->si_uid);
1165 retval = put_user(status, &infop->si_status);
1173 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1174 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1175 * the lock and this task is uninteresting. If we return nonzero, we have
1176 * released the lock and the system call should return.
1178 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1180 unsigned long state;
1181 int retval, status, traced;
1182 pid_t pid = task_pid_vnr(p);
1183 uid_t uid = __task_cred(p)->uid;
1184 struct siginfo __user *infop;
1186 if (!likely(wo->wo_flags & WEXITED))
1189 if (unlikely(wo->wo_flags & WNOWAIT)) {
1190 int exit_code = p->exit_code;
1194 read_unlock(&tasklist_lock);
1195 if ((exit_code & 0x7f) == 0) {
1197 status = exit_code >> 8;
1199 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1200 status = exit_code & 0x7f;
1202 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1206 * Try to move the task's state to DEAD
1207 * only one thread is allowed to do this:
1209 state = xchg(&p->exit_state, EXIT_DEAD);
1210 if (state != EXIT_ZOMBIE) {
1211 BUG_ON(state != EXIT_DEAD);
1215 traced = ptrace_reparented(p);
1217 * It can be ptraced but not reparented, check
1218 * !task_detached() to filter out sub-threads.
1220 if (likely(!traced) && likely(!task_detached(p))) {
1221 struct signal_struct *psig;
1222 struct signal_struct *sig;
1223 unsigned long maxrss;
1224 cputime_t tgutime, tgstime;
1227 * The resource counters for the group leader are in its
1228 * own task_struct. Those for dead threads in the group
1229 * are in its signal_struct, as are those for the child
1230 * processes it has previously reaped. All these
1231 * accumulate in the parent's signal_struct c* fields.
1233 * We don't bother to take a lock here to protect these
1234 * p->signal fields, because they are only touched by
1235 * __exit_signal, which runs with tasklist_lock
1236 * write-locked anyway, and so is excluded here. We do
1237 * need to protect the access to parent->signal fields,
1238 * as other threads in the parent group can be right
1239 * here reaping other children at the same time.
1241 * We use thread_group_times() to get times for the thread
1242 * group, which consolidates times for all threads in the
1243 * group including the group leader.
1245 thread_group_times(p, &tgutime, &tgstime);
1246 spin_lock_irq(&p->real_parent->sighand->siglock);
1247 psig = p->real_parent->signal;
1250 cputime_add(psig->cutime,
1251 cputime_add(tgutime,
1254 cputime_add(psig->cstime,
1255 cputime_add(tgstime,
1258 cputime_add(psig->cgtime,
1259 cputime_add(p->gtime,
1260 cputime_add(sig->gtime,
1263 p->min_flt + sig->min_flt + sig->cmin_flt;
1265 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1267 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1269 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1271 task_io_get_inblock(p) +
1272 sig->inblock + sig->cinblock;
1274 task_io_get_oublock(p) +
1275 sig->oublock + sig->coublock;
1276 maxrss = max(sig->maxrss, sig->cmaxrss);
1277 if (psig->cmaxrss < maxrss)
1278 psig->cmaxrss = maxrss;
1279 task_io_accounting_add(&psig->ioac, &p->ioac);
1280 task_io_accounting_add(&psig->ioac, &sig->ioac);
1281 spin_unlock_irq(&p->real_parent->sighand->siglock);
1285 * Now we are sure this task is interesting, and no other
1286 * thread can reap it because we set its state to EXIT_DEAD.
1288 read_unlock(&tasklist_lock);
1290 retval = wo->wo_rusage
1291 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1292 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1293 ? p->signal->group_exit_code : p->exit_code;
1294 if (!retval && wo->wo_stat)
1295 retval = put_user(status, wo->wo_stat);
1297 infop = wo->wo_info;
1298 if (!retval && infop)
1299 retval = put_user(SIGCHLD, &infop->si_signo);
1300 if (!retval && infop)
1301 retval = put_user(0, &infop->si_errno);
1302 if (!retval && infop) {
1305 if ((status & 0x7f) == 0) {
1309 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1312 retval = put_user((short)why, &infop->si_code);
1314 retval = put_user(status, &infop->si_status);
1316 if (!retval && infop)
1317 retval = put_user(pid, &infop->si_pid);
1318 if (!retval && infop)
1319 retval = put_user(uid, &infop->si_uid);
1324 write_lock_irq(&tasklist_lock);
1325 /* We dropped tasklist, ptracer could die and untrace */
1328 * If this is not a detached task, notify the parent.
1329 * If it's still not detached after that, don't release
1332 if (!task_detached(p)) {
1333 do_notify_parent(p, p->exit_signal);
1334 if (!task_detached(p)) {
1335 p->exit_state = EXIT_ZOMBIE;
1339 write_unlock_irq(&tasklist_lock);
1347 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1350 if (task_is_stopped_or_traced(p))
1351 return &p->exit_code;
1353 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1354 return &p->signal->group_exit_code;
1360 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1361 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1362 * the lock and this task is uninteresting. If we return nonzero, we have
1363 * released the lock and the system call should return.
1365 static int wait_task_stopped(struct wait_opts *wo,
1366 int ptrace, struct task_struct *p)
1368 struct siginfo __user *infop;
1369 int retval, exit_code, *p_code, why;
1370 uid_t uid = 0; /* unneeded, required by compiler */
1374 * Traditionally we see ptrace'd stopped tasks regardless of options.
1376 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1380 spin_lock_irq(&p->sighand->siglock);
1382 p_code = task_stopped_code(p, ptrace);
1383 if (unlikely(!p_code))
1386 exit_code = *p_code;
1390 if (!unlikely(wo->wo_flags & WNOWAIT))
1395 spin_unlock_irq(&p->sighand->siglock);
1400 * Now we are pretty sure this task is interesting.
1401 * Make sure it doesn't get reaped out from under us while we
1402 * give up the lock and then examine it below. We don't want to
1403 * keep holding onto the tasklist_lock while we call getrusage and
1404 * possibly take page faults for user memory.
1407 pid = task_pid_vnr(p);
1408 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1409 read_unlock(&tasklist_lock);
1411 if (unlikely(wo->wo_flags & WNOWAIT))
1412 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1414 retval = wo->wo_rusage
1415 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1416 if (!retval && wo->wo_stat)
1417 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1419 infop = wo->wo_info;
1420 if (!retval && infop)
1421 retval = put_user(SIGCHLD, &infop->si_signo);
1422 if (!retval && infop)
1423 retval = put_user(0, &infop->si_errno);
1424 if (!retval && infop)
1425 retval = put_user((short)why, &infop->si_code);
1426 if (!retval && infop)
1427 retval = put_user(exit_code, &infop->si_status);
1428 if (!retval && infop)
1429 retval = put_user(pid, &infop->si_pid);
1430 if (!retval && infop)
1431 retval = put_user(uid, &infop->si_uid);
1441 * Handle do_wait work for one task in a live, non-stopped state.
1442 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1443 * the lock and this task is uninteresting. If we return nonzero, we have
1444 * released the lock and the system call should return.
1446 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1452 if (!unlikely(wo->wo_flags & WCONTINUED))
1455 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1458 spin_lock_irq(&p->sighand->siglock);
1459 /* Re-check with the lock held. */
1460 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1461 spin_unlock_irq(&p->sighand->siglock);
1464 if (!unlikely(wo->wo_flags & WNOWAIT))
1465 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1467 spin_unlock_irq(&p->sighand->siglock);
1469 pid = task_pid_vnr(p);
1471 read_unlock(&tasklist_lock);
1474 retval = wo->wo_rusage
1475 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1477 if (!retval && wo->wo_stat)
1478 retval = put_user(0xffff, wo->wo_stat);
1482 retval = wait_noreap_copyout(wo, p, pid, uid,
1483 CLD_CONTINUED, SIGCONT);
1484 BUG_ON(retval == 0);
1491 * Consider @p for a wait by @parent.
1493 * -ECHILD should be in ->notask_error before the first call.
1494 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1495 * Returns zero if the search for a child should continue;
1496 * then ->notask_error is 0 if @p is an eligible child,
1497 * or another error from security_task_wait(), or still -ECHILD.
1499 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1500 struct task_struct *p)
1502 int ret = eligible_child(wo, p);
1506 ret = security_task_wait(p);
1507 if (unlikely(ret < 0)) {
1509 * If we have not yet seen any eligible child,
1510 * then let this error code replace -ECHILD.
1511 * A permission error will give the user a clue
1512 * to look for security policy problems, rather
1513 * than for mysterious wait bugs.
1515 if (wo->notask_error)
1516 wo->notask_error = ret;
1520 if (likely(!ptrace) && unlikely(task_ptrace(p))) {
1522 * This child is hidden by ptrace.
1523 * We aren't allowed to see it now, but eventually we will.
1525 wo->notask_error = 0;
1529 if (p->exit_state == EXIT_DEAD)
1533 * We don't reap group leaders with subthreads.
1535 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1536 return wait_task_zombie(wo, p);
1539 * It's stopped or running now, so it might
1540 * later continue, exit, or stop again.
1542 wo->notask_error = 0;
1544 if (task_stopped_code(p, ptrace))
1545 return wait_task_stopped(wo, ptrace, p);
1547 return wait_task_continued(wo, p);
1551 * Do the work of do_wait() for one thread in the group, @tsk.
1553 * -ECHILD should be in ->notask_error before the first call.
1554 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1555 * Returns zero if the search for a child should continue; then
1556 * ->notask_error is 0 if there were any eligible children,
1557 * or another error from security_task_wait(), or still -ECHILD.
1559 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1561 struct task_struct *p;
1563 list_for_each_entry(p, &tsk->children, sibling) {
1565 * Do not consider detached threads.
1567 if (!task_detached(p)) {
1568 int ret = wait_consider_task(wo, 0, p);
1577 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1579 struct task_struct *p;
1581 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1582 int ret = wait_consider_task(wo, 1, p);
1590 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1591 int sync, void *key)
1593 struct wait_opts *wo = container_of(wait, struct wait_opts,
1595 struct task_struct *p = key;
1597 if (!eligible_pid(wo, p))
1600 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1603 return default_wake_function(wait, mode, sync, key);
1606 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1608 __wake_up_sync_key(&parent->signal->wait_chldexit,
1609 TASK_INTERRUPTIBLE, 1, p);
1612 static long do_wait(struct wait_opts *wo)
1614 struct task_struct *tsk;
1617 trace_sched_process_wait(wo->wo_pid);
1619 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1620 wo->child_wait.private = current;
1621 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1624 * If there is nothing that can match our critiera just get out.
1625 * We will clear ->notask_error to zero if we see any child that
1626 * might later match our criteria, even if we are not able to reap
1629 wo->notask_error = -ECHILD;
1630 if ((wo->wo_type < PIDTYPE_MAX) &&
1631 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1634 set_current_state(TASK_INTERRUPTIBLE);
1635 read_lock(&tasklist_lock);
1638 retval = do_wait_thread(wo, tsk);
1642 retval = ptrace_do_wait(wo, tsk);
1646 if (wo->wo_flags & __WNOTHREAD)
1648 } while_each_thread(current, tsk);
1649 read_unlock(&tasklist_lock);
1652 retval = wo->notask_error;
1653 if (!retval && !(wo->wo_flags & WNOHANG)) {
1654 retval = -ERESTARTSYS;
1655 if (!signal_pending(current)) {
1661 __set_current_state(TASK_RUNNING);
1662 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1666 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1667 infop, int, options, struct rusage __user *, ru)
1669 struct wait_opts wo;
1670 struct pid *pid = NULL;
1674 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1676 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1689 type = PIDTYPE_PGID;
1697 if (type < PIDTYPE_MAX)
1698 pid = find_get_pid(upid);
1702 wo.wo_flags = options;
1712 * For a WNOHANG return, clear out all the fields
1713 * we would set so the user can easily tell the
1717 ret = put_user(0, &infop->si_signo);
1719 ret = put_user(0, &infop->si_errno);
1721 ret = put_user(0, &infop->si_code);
1723 ret = put_user(0, &infop->si_pid);
1725 ret = put_user(0, &infop->si_uid);
1727 ret = put_user(0, &infop->si_status);
1732 /* avoid REGPARM breakage on x86: */
1733 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1737 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1738 int, options, struct rusage __user *, ru)
1740 struct wait_opts wo;
1741 struct pid *pid = NULL;
1745 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1746 __WNOTHREAD|__WCLONE|__WALL))
1751 else if (upid < 0) {
1752 type = PIDTYPE_PGID;
1753 pid = find_get_pid(-upid);
1754 } else if (upid == 0) {
1755 type = PIDTYPE_PGID;
1756 pid = get_task_pid(current, PIDTYPE_PGID);
1757 } else /* upid > 0 */ {
1759 pid = find_get_pid(upid);
1764 wo.wo_flags = options | WEXITED;
1766 wo.wo_stat = stat_addr;
1771 /* avoid REGPARM breakage on x86: */
1772 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1776 #ifdef __ARCH_WANT_SYS_WAITPID
1779 * sys_waitpid() remains for compatibility. waitpid() should be
1780 * implemented by calling sys_wait4() from libc.a.
1782 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1784 return sys_wait4(pid, stat_addr, options, NULL);