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>
54 #include <asm/uaccess.h>
55 #include <asm/unistd.h>
56 #include <asm/pgtable.h>
57 #include <asm/mmu_context.h>
59 static void exit_mm(struct task_struct * tsk);
61 static void __unhash_process(struct task_struct *p, bool group_dead)
64 detach_pid(p, PIDTYPE_PID);
66 detach_pid(p, PIDTYPE_PGID);
67 detach_pid(p, PIDTYPE_SID);
69 list_del_rcu(&p->tasks);
70 list_del_init(&p->sibling);
71 __get_cpu_var(process_counts)--;
73 list_del_rcu(&p->thread_group);
77 * This function expects the tasklist_lock write-locked.
79 static void __exit_signal(struct task_struct *tsk)
81 struct signal_struct *sig = tsk->signal;
82 bool group_dead = thread_group_leader(tsk);
83 struct sighand_struct *sighand;
84 struct tty_struct *uninitialized_var(tty);
86 sighand = rcu_dereference_check(tsk->sighand,
87 rcu_read_lock_held() ||
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 #ifdef CONFIG_PERF_EVENTS
161 WARN_ON_ONCE(tsk->perf_event_ctxp);
163 trace_sched_process_free(tsk);
164 put_task_struct(tsk);
168 void release_task(struct task_struct * p)
170 struct task_struct *leader;
173 tracehook_prepare_release_task(p);
174 /* don't need to get the RCU readlock here - the process is dead and
175 * can't be modifying its own credentials. But shut RCU-lockdep up */
177 atomic_dec(&__task_cred(p)->user->processes);
182 write_lock_irq(&tasklist_lock);
183 tracehook_finish_release_task(p);
187 * If we are the last non-leader member of the thread
188 * group, and the leader is zombie, then notify the
189 * group leader's parent process. (if it wants notification.)
192 leader = p->group_leader;
193 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
194 BUG_ON(task_detached(leader));
195 do_notify_parent(leader, leader->exit_signal);
197 * If we were the last child thread and the leader has
198 * exited already, and the leader's parent ignores SIGCHLD,
199 * then we are the one who should release the leader.
201 * do_notify_parent() will have marked it self-reaping in
204 zap_leader = task_detached(leader);
207 * This maintains the invariant that release_task()
208 * only runs on a task in EXIT_DEAD, just for sanity.
211 leader->exit_state = EXIT_DEAD;
214 write_unlock_irq(&tasklist_lock);
216 call_rcu(&p->rcu, delayed_put_task_struct);
219 if (unlikely(zap_leader))
224 * This checks not only the pgrp, but falls back on the pid if no
225 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
228 * The caller must hold rcu lock or the tasklist lock.
230 struct pid *session_of_pgrp(struct pid *pgrp)
232 struct task_struct *p;
233 struct pid *sid = NULL;
235 p = pid_task(pgrp, PIDTYPE_PGID);
237 p = pid_task(pgrp, PIDTYPE_PID);
239 sid = task_session(p);
245 * Determine if a process group is "orphaned", according to the POSIX
246 * definition in 2.2.2.52. Orphaned process groups are not to be affected
247 * by terminal-generated stop signals. Newly orphaned process groups are
248 * to receive a SIGHUP and a SIGCONT.
250 * "I ask you, have you ever known what it is to be an orphan?"
252 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
254 struct task_struct *p;
256 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
257 if ((p == ignored_task) ||
258 (p->exit_state && thread_group_empty(p)) ||
259 is_global_init(p->real_parent))
262 if (task_pgrp(p->real_parent) != pgrp &&
263 task_session(p->real_parent) == task_session(p))
265 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
270 int is_current_pgrp_orphaned(void)
274 read_lock(&tasklist_lock);
275 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
276 read_unlock(&tasklist_lock);
281 static int has_stopped_jobs(struct pid *pgrp)
284 struct task_struct *p;
286 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
287 if (!task_is_stopped(p))
291 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
296 * Check to see if any process groups have become orphaned as
297 * a result of our exiting, and if they have any stopped jobs,
298 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
301 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
303 struct pid *pgrp = task_pgrp(tsk);
304 struct task_struct *ignored_task = tsk;
307 /* exit: our father is in a different pgrp than
308 * we are and we were the only connection outside.
310 parent = tsk->real_parent;
312 /* reparent: our child is in a different pgrp than
313 * we are, and it was the only connection outside.
317 if (task_pgrp(parent) != pgrp &&
318 task_session(parent) == task_session(tsk) &&
319 will_become_orphaned_pgrp(pgrp, ignored_task) &&
320 has_stopped_jobs(pgrp)) {
321 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
322 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
327 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
329 * If a kernel thread is launched as a result of a system call, or if
330 * it ever exits, it should generally reparent itself to kthreadd so it
331 * isn't in the way of other processes and is correctly cleaned up on exit.
333 * The various task state such as scheduling policy and priority may have
334 * been inherited from a user process, so we reset them to sane values here.
336 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
338 static void reparent_to_kthreadd(void)
340 write_lock_irq(&tasklist_lock);
342 ptrace_unlink(current);
343 /* Reparent to init */
344 current->real_parent = current->parent = kthreadd_task;
345 list_move_tail(¤t->sibling, ¤t->real_parent->children);
347 /* Set the exit signal to SIGCHLD so we signal init on exit */
348 current->exit_signal = SIGCHLD;
350 if (task_nice(current) < 0)
351 set_user_nice(current, 0);
355 memcpy(current->signal->rlim, init_task.signal->rlim,
356 sizeof(current->signal->rlim));
358 atomic_inc(&init_cred.usage);
359 commit_creds(&init_cred);
360 write_unlock_irq(&tasklist_lock);
363 void __set_special_pids(struct pid *pid)
365 struct task_struct *curr = current->group_leader;
367 if (task_session(curr) != pid)
368 change_pid(curr, PIDTYPE_SID, pid);
370 if (task_pgrp(curr) != pid)
371 change_pid(curr, PIDTYPE_PGID, pid);
374 static void set_special_pids(struct pid *pid)
376 write_lock_irq(&tasklist_lock);
377 __set_special_pids(pid);
378 write_unlock_irq(&tasklist_lock);
382 * Let kernel threads use this to say that they allow a certain signal.
383 * Must not be used if kthread was cloned with CLONE_SIGHAND.
385 int allow_signal(int sig)
387 if (!valid_signal(sig) || sig < 1)
390 spin_lock_irq(¤t->sighand->siglock);
391 /* This is only needed for daemonize()'ed kthreads */
392 sigdelset(¤t->blocked, sig);
394 * Kernel threads handle their own signals. Let the signal code
395 * know it'll be handled, so that they don't get converted to
396 * SIGKILL or just silently dropped.
398 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
400 spin_unlock_irq(¤t->sighand->siglock);
404 EXPORT_SYMBOL(allow_signal);
406 int disallow_signal(int sig)
408 if (!valid_signal(sig) || sig < 1)
411 spin_lock_irq(¤t->sighand->siglock);
412 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
414 spin_unlock_irq(¤t->sighand->siglock);
418 EXPORT_SYMBOL(disallow_signal);
421 * Put all the gunge required to become a kernel thread without
422 * attached user resources in one place where it belongs.
425 void daemonize(const char *name, ...)
430 va_start(args, name);
431 vsnprintf(current->comm, sizeof(current->comm), name, args);
435 * If we were started as result of loading a module, close all of the
436 * user space pages. We don't need them, and if we didn't close them
437 * they would be locked into memory.
441 * We don't want to have TIF_FREEZE set if the system-wide hibernation
442 * or suspend transition begins right now.
444 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
446 if (current->nsproxy != &init_nsproxy) {
447 get_nsproxy(&init_nsproxy);
448 switch_task_namespaces(current, &init_nsproxy);
450 set_special_pids(&init_struct_pid);
451 proc_clear_tty(current);
453 /* Block and flush all signals */
454 sigfillset(&blocked);
455 sigprocmask(SIG_BLOCK, &blocked, NULL);
456 flush_signals(current);
458 /* Become as one with the init task */
460 daemonize_fs_struct();
462 current->files = init_task.files;
463 atomic_inc(¤t->files->count);
465 reparent_to_kthreadd();
468 EXPORT_SYMBOL(daemonize);
470 static void close_files(struct files_struct * files)
478 * It is safe to dereference the fd table without RCU or
479 * ->file_lock because this is the last reference to the
480 * files structure. But use RCU to shut RCU-lockdep up.
483 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.
531 fdt = files_fdtable(files);
532 if (fdt != &files->fdtab)
533 kmem_cache_free(files_cachep, files);
539 void reset_files_struct(struct files_struct *files)
541 struct task_struct *tsk = current;
542 struct files_struct *old;
548 put_files_struct(old);
551 void exit_files(struct task_struct *tsk)
553 struct files_struct * files = tsk->files;
559 put_files_struct(files);
563 #ifdef CONFIG_MM_OWNER
565 * Task p is exiting and it owned mm, lets find a new owner for it
568 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
571 * If there are other users of the mm and the owner (us) is exiting
572 * we need to find a new owner to take on the responsibility.
574 if (atomic_read(&mm->mm_users) <= 1)
581 void mm_update_next_owner(struct mm_struct *mm)
583 struct task_struct *c, *g, *p = current;
586 if (!mm_need_new_owner(mm, p))
589 read_lock(&tasklist_lock);
591 * Search in the children
593 list_for_each_entry(c, &p->children, sibling) {
595 goto assign_new_owner;
599 * Search in the siblings
601 list_for_each_entry(c, &p->real_parent->children, sibling) {
603 goto assign_new_owner;
607 * Search through everything else. We should not get
610 do_each_thread(g, c) {
612 goto assign_new_owner;
613 } while_each_thread(g, c);
615 read_unlock(&tasklist_lock);
617 * We found no owner yet mm_users > 1: this implies that we are
618 * most likely racing with swapoff (try_to_unuse()) or /proc or
619 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
628 * The task_lock protects c->mm from changing.
629 * We always want mm->owner->mm == mm
633 * Delay read_unlock() till we have the task_lock()
634 * to ensure that c does not slip away underneath us
636 read_unlock(&tasklist_lock);
646 #endif /* CONFIG_MM_OWNER */
649 * Turn us into a lazy TLB process if we
652 static void exit_mm(struct task_struct * tsk)
654 struct mm_struct *mm = tsk->mm;
655 struct core_state *core_state;
661 * Serialize with any possible pending coredump.
662 * We must hold mmap_sem around checking core_state
663 * and clearing tsk->mm. The core-inducing thread
664 * will increment ->nr_threads for each thread in the
665 * group with ->mm != NULL.
667 down_read(&mm->mmap_sem);
668 core_state = mm->core_state;
670 struct core_thread self;
671 up_read(&mm->mmap_sem);
674 self.next = xchg(&core_state->dumper.next, &self);
676 * Implies mb(), the result of xchg() must be visible
677 * to core_state->dumper.
679 if (atomic_dec_and_test(&core_state->nr_threads))
680 complete(&core_state->startup);
683 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
684 if (!self.task) /* see coredump_finish() */
688 __set_task_state(tsk, TASK_RUNNING);
689 down_read(&mm->mmap_sem);
691 atomic_inc(&mm->mm_count);
692 BUG_ON(mm != tsk->active_mm);
693 /* more a memory barrier than a real lock */
696 up_read(&mm->mmap_sem);
697 enter_lazy_tlb(mm, current);
698 /* We don't want this task to be frozen prematurely */
699 clear_freeze_flag(tsk);
701 mm_update_next_owner(mm);
706 * When we die, we re-parent all our children.
707 * Try to give them to another thread in our thread
708 * group, and if no such member exists, give it to
709 * the child reaper process (ie "init") in our pid
712 static struct task_struct *find_new_reaper(struct task_struct *father)
714 struct pid_namespace *pid_ns = task_active_pid_ns(father);
715 struct task_struct *thread;
718 while_each_thread(father, thread) {
719 if (thread->flags & PF_EXITING)
721 if (unlikely(pid_ns->child_reaper == father))
722 pid_ns->child_reaper = thread;
726 if (unlikely(pid_ns->child_reaper == father)) {
727 write_unlock_irq(&tasklist_lock);
728 if (unlikely(pid_ns == &init_pid_ns))
729 panic("Attempted to kill init!");
731 zap_pid_ns_processes(pid_ns);
732 write_lock_irq(&tasklist_lock);
734 * We can not clear ->child_reaper or leave it alone.
735 * There may by stealth EXIT_DEAD tasks on ->children,
736 * forget_original_parent() must move them somewhere.
738 pid_ns->child_reaper = init_pid_ns.child_reaper;
741 return pid_ns->child_reaper;
745 * Any that need to be release_task'd are put on the @dead list.
747 static void reparent_leader(struct task_struct *father, struct task_struct *p,
748 struct list_head *dead)
750 list_move_tail(&p->sibling, &p->real_parent->children);
752 if (task_detached(p))
755 * If this is a threaded reparent there is no need to
756 * notify anyone anything has happened.
758 if (same_thread_group(p->real_parent, father))
761 /* We don't want people slaying init. */
762 p->exit_signal = SIGCHLD;
764 /* If it has exited notify the new parent about this child's death. */
765 if (!task_ptrace(p) &&
766 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
767 do_notify_parent(p, p->exit_signal);
768 if (task_detached(p)) {
769 p->exit_state = EXIT_DEAD;
770 list_move_tail(&p->sibling, dead);
774 kill_orphaned_pgrp(p, father);
777 static void forget_original_parent(struct task_struct *father)
779 struct task_struct *p, *n, *reaper;
780 LIST_HEAD(dead_children);
782 write_lock_irq(&tasklist_lock);
784 * Note that exit_ptrace() and find_new_reaper() might
785 * drop tasklist_lock and reacquire it.
788 reaper = find_new_reaper(father);
790 list_for_each_entry_safe(p, n, &father->children, sibling) {
791 struct task_struct *t = p;
793 t->real_parent = reaper;
794 if (t->parent == father) {
795 BUG_ON(task_ptrace(t));
796 t->parent = t->real_parent;
798 if (t->pdeath_signal)
799 group_send_sig_info(t->pdeath_signal,
801 } while_each_thread(p, t);
802 reparent_leader(father, p, &dead_children);
804 write_unlock_irq(&tasklist_lock);
806 BUG_ON(!list_empty(&father->children));
808 list_for_each_entry_safe(p, n, &dead_children, sibling) {
809 list_del_init(&p->sibling);
815 * Send signals to all our closest relatives so that they know
816 * to properly mourn us..
818 static void exit_notify(struct task_struct *tsk, int group_dead)
824 * This does two things:
826 * A. Make init inherit all the child processes
827 * B. Check to see if any process groups have become orphaned
828 * as a result of our exiting, and if they have any stopped
829 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
831 forget_original_parent(tsk);
832 exit_task_namespaces(tsk);
834 write_lock_irq(&tasklist_lock);
836 kill_orphaned_pgrp(tsk->group_leader, NULL);
838 /* Let father know we died
840 * Thread signals are configurable, but you aren't going to use
841 * that to send signals to arbitary processes.
842 * That stops right now.
844 * If the parent exec id doesn't match the exec id we saved
845 * when we started then we know the parent has changed security
848 * If our self_exec id doesn't match our parent_exec_id then
849 * we have changed execution domain as these two values started
850 * the same after a fork.
852 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
853 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
854 tsk->self_exec_id != tsk->parent_exec_id))
855 tsk->exit_signal = SIGCHLD;
857 signal = tracehook_notify_death(tsk, &cookie, group_dead);
859 signal = do_notify_parent(tsk, signal);
861 tsk->exit_state = signal == DEATH_REAP ? 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 tracehook_report_death(tsk, signal, cookie, group_dead);
870 /* If the process is dead, release it - nobody will wait for it */
871 if (signal == DEATH_REAP)
875 #ifdef CONFIG_DEBUG_STACK_USAGE
876 static void check_stack_usage(void)
878 static DEFINE_SPINLOCK(low_water_lock);
879 static int lowest_to_date = THREAD_SIZE;
882 free = stack_not_used(current);
884 if (free >= lowest_to_date)
887 spin_lock(&low_water_lock);
888 if (free < lowest_to_date) {
889 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
891 current->comm, free);
892 lowest_to_date = free;
894 spin_unlock(&low_water_lock);
897 static inline void check_stack_usage(void) {}
900 NORET_TYPE void do_exit(long code)
902 struct task_struct *tsk = current;
905 profile_task_exit(tsk);
907 WARN_ON(atomic_read(&tsk->fs_excl));
909 if (unlikely(in_interrupt()))
910 panic("Aiee, killing interrupt handler!");
911 if (unlikely(!tsk->pid))
912 panic("Attempted to kill the idle task!");
915 * If do_exit is called because this processes oopsed, it's possible
916 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
917 * continuing. Amongst other possible reasons, this is to prevent
918 * mm_release()->clear_child_tid() from writing to a user-controlled
923 tracehook_report_exit(&code);
925 validate_creds_for_do_exit(tsk);
928 * We're taking recursive faults here in do_exit. Safest is to just
929 * leave this task alone and wait for reboot.
931 if (unlikely(tsk->flags & PF_EXITING)) {
933 "Fixing recursive fault but reboot is needed!\n");
935 * We can do this unlocked here. The futex code uses
936 * this flag just to verify whether the pi state
937 * cleanup has been done or not. In the worst case it
938 * loops once more. We pretend that the cleanup was
939 * done as there is no way to return. Either the
940 * OWNER_DIED bit is set by now or we push the blocked
941 * task into the wait for ever nirwana as well.
943 tsk->flags |= PF_EXITPIDONE;
944 set_current_state(TASK_UNINTERRUPTIBLE);
950 exit_signals(tsk); /* sets PF_EXITING */
952 * tsk->flags are checked in the futex code to protect against
953 * an exiting task cleaning up the robust pi futexes.
956 raw_spin_unlock_wait(&tsk->pi_lock);
958 if (unlikely(in_atomic()))
959 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
960 current->comm, task_pid_nr(current),
963 acct_update_integrals(tsk);
964 /* sync mm's RSS info before statistics gathering */
966 sync_mm_rss(tsk, tsk->mm);
967 group_dead = atomic_dec_and_test(&tsk->signal->live);
969 hrtimer_cancel(&tsk->signal->real_timer);
970 exit_itimers(tsk->signal);
972 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
974 acct_collect(code, group_dead);
977 if (unlikely(tsk->audit_context))
980 tsk->exit_code = code;
981 taskstats_exit(tsk, group_dead);
987 trace_sched_process_exit(tsk);
997 disassociate_ctty(1);
999 module_put(task_thread_info(tsk)->exec_domain->module);
1001 proc_exit_connector(tsk);
1004 * FIXME: do that only when needed, using sched_exit tracepoint
1006 flush_ptrace_hw_breakpoint(tsk);
1008 * Flush inherited counters to the parent - before the parent
1009 * gets woken up by child-exit notifications.
1011 perf_event_exit_task(tsk);
1013 exit_notify(tsk, group_dead);
1016 mpol_put(tsk->mempolicy);
1017 tsk->mempolicy = NULL;
1021 if (unlikely(current->pi_state_cache))
1022 kfree(current->pi_state_cache);
1025 * Make sure we are holding no locks:
1027 debug_check_no_locks_held(tsk);
1029 * We can do this unlocked here. The futex code uses this flag
1030 * just to verify whether the pi state cleanup has been done
1031 * or not. In the worst case it loops once more.
1033 tsk->flags |= PF_EXITPIDONE;
1035 if (tsk->io_context)
1036 exit_io_context(tsk);
1038 if (tsk->splice_pipe)
1039 __free_pipe_info(tsk->splice_pipe);
1041 validate_creds_for_do_exit(tsk);
1045 /* causes final put_task_struct in finish_task_switch(). */
1046 tsk->state = TASK_DEAD;
1049 /* Avoid "noreturn function does return". */
1051 cpu_relax(); /* For when BUG is null */
1054 EXPORT_SYMBOL_GPL(do_exit);
1056 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1064 EXPORT_SYMBOL(complete_and_exit);
1066 SYSCALL_DEFINE1(exit, int, error_code)
1068 do_exit((error_code&0xff)<<8);
1072 * Take down every thread in the group. This is called by fatal signals
1073 * as well as by sys_exit_group (below).
1076 do_group_exit(int exit_code)
1078 struct signal_struct *sig = current->signal;
1080 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1082 if (signal_group_exit(sig))
1083 exit_code = sig->group_exit_code;
1084 else if (!thread_group_empty(current)) {
1085 struct sighand_struct *const sighand = current->sighand;
1086 spin_lock_irq(&sighand->siglock);
1087 if (signal_group_exit(sig))
1088 /* Another thread got here before we took the lock. */
1089 exit_code = sig->group_exit_code;
1091 sig->group_exit_code = exit_code;
1092 sig->flags = SIGNAL_GROUP_EXIT;
1093 zap_other_threads(current);
1095 spin_unlock_irq(&sighand->siglock);
1103 * this kills every thread in the thread group. Note that any externally
1104 * wait4()-ing process will get the correct exit code - even if this
1105 * thread is not the thread group leader.
1107 SYSCALL_DEFINE1(exit_group, int, error_code)
1109 do_group_exit((error_code & 0xff) << 8);
1115 enum pid_type wo_type;
1119 struct siginfo __user *wo_info;
1120 int __user *wo_stat;
1121 struct rusage __user *wo_rusage;
1123 wait_queue_t child_wait;
1128 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1130 if (type != PIDTYPE_PID)
1131 task = task->group_leader;
1132 return task->pids[type].pid;
1135 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1137 return wo->wo_type == PIDTYPE_MAX ||
1138 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1141 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1143 if (!eligible_pid(wo, p))
1145 /* Wait for all children (clone and not) if __WALL is set;
1146 * otherwise, wait for clone children *only* if __WCLONE is
1147 * set; otherwise, wait for non-clone children *only*. (Note:
1148 * A "clone" child here is one that reports to its parent
1149 * using a signal other than SIGCHLD.) */
1150 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1151 && !(wo->wo_flags & __WALL))
1157 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1158 pid_t pid, uid_t uid, int why, int status)
1160 struct siginfo __user *infop;
1161 int retval = wo->wo_rusage
1162 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1165 infop = wo->wo_info;
1168 retval = put_user(SIGCHLD, &infop->si_signo);
1170 retval = put_user(0, &infop->si_errno);
1172 retval = put_user((short)why, &infop->si_code);
1174 retval = put_user(pid, &infop->si_pid);
1176 retval = put_user(uid, &infop->si_uid);
1178 retval = put_user(status, &infop->si_status);
1186 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1187 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1188 * the lock and this task is uninteresting. If we return nonzero, we have
1189 * released the lock and the system call should return.
1191 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1193 unsigned long state;
1194 int retval, status, traced;
1195 pid_t pid = task_pid_vnr(p);
1196 uid_t uid = __task_cred(p)->uid;
1197 struct siginfo __user *infop;
1199 if (!likely(wo->wo_flags & WEXITED))
1202 if (unlikely(wo->wo_flags & WNOWAIT)) {
1203 int exit_code = p->exit_code;
1207 read_unlock(&tasklist_lock);
1208 if ((exit_code & 0x7f) == 0) {
1210 status = exit_code >> 8;
1212 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1213 status = exit_code & 0x7f;
1215 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1219 * Try to move the task's state to DEAD
1220 * only one thread is allowed to do this:
1222 state = xchg(&p->exit_state, EXIT_DEAD);
1223 if (state != EXIT_ZOMBIE) {
1224 BUG_ON(state != EXIT_DEAD);
1228 traced = ptrace_reparented(p);
1230 * It can be ptraced but not reparented, check
1231 * !task_detached() to filter out sub-threads.
1233 if (likely(!traced) && likely(!task_detached(p))) {
1234 struct signal_struct *psig;
1235 struct signal_struct *sig;
1236 unsigned long maxrss;
1237 cputime_t tgutime, tgstime;
1240 * The resource counters for the group leader are in its
1241 * own task_struct. Those for dead threads in the group
1242 * are in its signal_struct, as are those for the child
1243 * processes it has previously reaped. All these
1244 * accumulate in the parent's signal_struct c* fields.
1246 * We don't bother to take a lock here to protect these
1247 * p->signal fields, because they are only touched by
1248 * __exit_signal, which runs with tasklist_lock
1249 * write-locked anyway, and so is excluded here. We do
1250 * need to protect the access to parent->signal fields,
1251 * as other threads in the parent group can be right
1252 * here reaping other children at the same time.
1254 * We use thread_group_times() to get times for the thread
1255 * group, which consolidates times for all threads in the
1256 * group including the group leader.
1258 thread_group_times(p, &tgutime, &tgstime);
1259 spin_lock_irq(&p->real_parent->sighand->siglock);
1260 psig = p->real_parent->signal;
1263 cputime_add(psig->cutime,
1264 cputime_add(tgutime,
1267 cputime_add(psig->cstime,
1268 cputime_add(tgstime,
1271 cputime_add(psig->cgtime,
1272 cputime_add(p->gtime,
1273 cputime_add(sig->gtime,
1276 p->min_flt + sig->min_flt + sig->cmin_flt;
1278 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1280 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1282 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1284 task_io_get_inblock(p) +
1285 sig->inblock + sig->cinblock;
1287 task_io_get_oublock(p) +
1288 sig->oublock + sig->coublock;
1289 maxrss = max(sig->maxrss, sig->cmaxrss);
1290 if (psig->cmaxrss < maxrss)
1291 psig->cmaxrss = maxrss;
1292 task_io_accounting_add(&psig->ioac, &p->ioac);
1293 task_io_accounting_add(&psig->ioac, &sig->ioac);
1294 spin_unlock_irq(&p->real_parent->sighand->siglock);
1298 * Now we are sure this task is interesting, and no other
1299 * thread can reap it because we set its state to EXIT_DEAD.
1301 read_unlock(&tasklist_lock);
1303 retval = wo->wo_rusage
1304 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1305 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1306 ? p->signal->group_exit_code : p->exit_code;
1307 if (!retval && wo->wo_stat)
1308 retval = put_user(status, wo->wo_stat);
1310 infop = wo->wo_info;
1311 if (!retval && infop)
1312 retval = put_user(SIGCHLD, &infop->si_signo);
1313 if (!retval && infop)
1314 retval = put_user(0, &infop->si_errno);
1315 if (!retval && infop) {
1318 if ((status & 0x7f) == 0) {
1322 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1325 retval = put_user((short)why, &infop->si_code);
1327 retval = put_user(status, &infop->si_status);
1329 if (!retval && infop)
1330 retval = put_user(pid, &infop->si_pid);
1331 if (!retval && infop)
1332 retval = put_user(uid, &infop->si_uid);
1337 write_lock_irq(&tasklist_lock);
1338 /* We dropped tasklist, ptracer could die and untrace */
1341 * If this is not a detached task, notify the parent.
1342 * If it's still not detached after that, don't release
1345 if (!task_detached(p)) {
1346 do_notify_parent(p, p->exit_signal);
1347 if (!task_detached(p)) {
1348 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 return &p->exit_code;
1366 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1367 return &p->signal->group_exit_code;
1373 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1374 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1375 * the lock and this task is uninteresting. If we return nonzero, we have
1376 * released the lock and the system call should return.
1378 static int wait_task_stopped(struct wait_opts *wo,
1379 int ptrace, struct task_struct *p)
1381 struct siginfo __user *infop;
1382 int retval, exit_code, *p_code, why;
1383 uid_t uid = 0; /* unneeded, required by compiler */
1387 * Traditionally we see ptrace'd stopped tasks regardless of options.
1389 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1393 spin_lock_irq(&p->sighand->siglock);
1395 p_code = task_stopped_code(p, ptrace);
1396 if (unlikely(!p_code))
1399 exit_code = *p_code;
1403 if (!unlikely(wo->wo_flags & WNOWAIT))
1408 spin_unlock_irq(&p->sighand->siglock);
1413 * Now we are pretty sure this task is interesting.
1414 * Make sure it doesn't get reaped out from under us while we
1415 * give up the lock and then examine it below. We don't want to
1416 * keep holding onto the tasklist_lock while we call getrusage and
1417 * possibly take page faults for user memory.
1420 pid = task_pid_vnr(p);
1421 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1422 read_unlock(&tasklist_lock);
1424 if (unlikely(wo->wo_flags & WNOWAIT))
1425 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1427 retval = wo->wo_rusage
1428 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1429 if (!retval && wo->wo_stat)
1430 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1432 infop = wo->wo_info;
1433 if (!retval && infop)
1434 retval = put_user(SIGCHLD, &infop->si_signo);
1435 if (!retval && infop)
1436 retval = put_user(0, &infop->si_errno);
1437 if (!retval && infop)
1438 retval = put_user((short)why, &infop->si_code);
1439 if (!retval && infop)
1440 retval = put_user(exit_code, &infop->si_status);
1441 if (!retval && infop)
1442 retval = put_user(pid, &infop->si_pid);
1443 if (!retval && infop)
1444 retval = put_user(uid, &infop->si_uid);
1454 * Handle do_wait work for one task in a live, non-stopped state.
1455 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1456 * the lock and this task is uninteresting. If we return nonzero, we have
1457 * released the lock and the system call should return.
1459 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1465 if (!unlikely(wo->wo_flags & WCONTINUED))
1468 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1471 spin_lock_irq(&p->sighand->siglock);
1472 /* Re-check with the lock held. */
1473 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1474 spin_unlock_irq(&p->sighand->siglock);
1477 if (!unlikely(wo->wo_flags & WNOWAIT))
1478 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1480 spin_unlock_irq(&p->sighand->siglock);
1482 pid = task_pid_vnr(p);
1484 read_unlock(&tasklist_lock);
1487 retval = wo->wo_rusage
1488 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1490 if (!retval && wo->wo_stat)
1491 retval = put_user(0xffff, wo->wo_stat);
1495 retval = wait_noreap_copyout(wo, p, pid, uid,
1496 CLD_CONTINUED, SIGCONT);
1497 BUG_ON(retval == 0);
1504 * Consider @p for a wait by @parent.
1506 * -ECHILD should be in ->notask_error before the first call.
1507 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1508 * Returns zero if the search for a child should continue;
1509 * then ->notask_error is 0 if @p is an eligible child,
1510 * or another error from security_task_wait(), or still -ECHILD.
1512 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1513 struct task_struct *p)
1515 int ret = eligible_child(wo, p);
1519 ret = security_task_wait(p);
1520 if (unlikely(ret < 0)) {
1522 * If we have not yet seen any eligible child,
1523 * then let this error code replace -ECHILD.
1524 * A permission error will give the user a clue
1525 * to look for security policy problems, rather
1526 * than for mysterious wait bugs.
1528 if (wo->notask_error)
1529 wo->notask_error = ret;
1533 if (likely(!ptrace) && unlikely(task_ptrace(p))) {
1535 * This child is hidden by ptrace.
1536 * We aren't allowed to see it now, but eventually we will.
1538 wo->notask_error = 0;
1542 if (p->exit_state == EXIT_DEAD)
1546 * We don't reap group leaders with subthreads.
1548 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1549 return wait_task_zombie(wo, p);
1552 * It's stopped or running now, so it might
1553 * later continue, exit, or stop again.
1555 wo->notask_error = 0;
1557 if (task_stopped_code(p, ptrace))
1558 return wait_task_stopped(wo, ptrace, p);
1560 return wait_task_continued(wo, p);
1564 * Do the work of do_wait() for one thread in the group, @tsk.
1566 * -ECHILD should be in ->notask_error before the first call.
1567 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1568 * Returns zero if the search for a child should continue; then
1569 * ->notask_error is 0 if there were any eligible children,
1570 * or another error from security_task_wait(), or still -ECHILD.
1572 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1574 struct task_struct *p;
1576 list_for_each_entry(p, &tsk->children, sibling) {
1577 int ret = wait_consider_task(wo, 0, p);
1585 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1587 struct task_struct *p;
1589 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1590 int ret = wait_consider_task(wo, 1, p);
1598 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1599 int sync, void *key)
1601 struct wait_opts *wo = container_of(wait, struct wait_opts,
1603 struct task_struct *p = key;
1605 if (!eligible_pid(wo, p))
1608 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1611 return default_wake_function(wait, mode, sync, key);
1614 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1616 __wake_up_sync_key(&parent->signal->wait_chldexit,
1617 TASK_INTERRUPTIBLE, 1, p);
1620 static long do_wait(struct wait_opts *wo)
1622 struct task_struct *tsk;
1625 trace_sched_process_wait(wo->wo_pid);
1627 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1628 wo->child_wait.private = current;
1629 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1632 * If there is nothing that can match our critiera just get out.
1633 * We will clear ->notask_error to zero if we see any child that
1634 * might later match our criteria, even if we are not able to reap
1637 wo->notask_error = -ECHILD;
1638 if ((wo->wo_type < PIDTYPE_MAX) &&
1639 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1642 set_current_state(TASK_INTERRUPTIBLE);
1643 read_lock(&tasklist_lock);
1646 retval = do_wait_thread(wo, tsk);
1650 retval = ptrace_do_wait(wo, tsk);
1654 if (wo->wo_flags & __WNOTHREAD)
1656 } while_each_thread(current, tsk);
1657 read_unlock(&tasklist_lock);
1660 retval = wo->notask_error;
1661 if (!retval && !(wo->wo_flags & WNOHANG)) {
1662 retval = -ERESTARTSYS;
1663 if (!signal_pending(current)) {
1669 __set_current_state(TASK_RUNNING);
1670 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1674 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1675 infop, int, options, struct rusage __user *, ru)
1677 struct wait_opts wo;
1678 struct pid *pid = NULL;
1682 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1684 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1697 type = PIDTYPE_PGID;
1705 if (type < PIDTYPE_MAX)
1706 pid = find_get_pid(upid);
1710 wo.wo_flags = options;
1720 * For a WNOHANG return, clear out all the fields
1721 * we would set so the user can easily tell the
1725 ret = put_user(0, &infop->si_signo);
1727 ret = put_user(0, &infop->si_errno);
1729 ret = put_user(0, &infop->si_code);
1731 ret = put_user(0, &infop->si_pid);
1733 ret = put_user(0, &infop->si_uid);
1735 ret = put_user(0, &infop->si_status);
1740 /* avoid REGPARM breakage on x86: */
1741 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1745 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1746 int, options, struct rusage __user *, ru)
1748 struct wait_opts wo;
1749 struct pid *pid = NULL;
1753 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1754 __WNOTHREAD|__WCLONE|__WALL))
1759 else if (upid < 0) {
1760 type = PIDTYPE_PGID;
1761 pid = find_get_pid(-upid);
1762 } else if (upid == 0) {
1763 type = PIDTYPE_PGID;
1764 pid = get_task_pid(current, PIDTYPE_PGID);
1765 } else /* upid > 0 */ {
1767 pid = find_get_pid(upid);
1772 wo.wo_flags = options | WEXITED;
1774 wo.wo_stat = stat_addr;
1779 /* avoid REGPARM breakage on x86: */
1780 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1784 #ifdef __ARCH_WANT_SYS_WAITPID
1787 * sys_waitpid() remains for compatibility. waitpid() should be
1788 * implemented by calling sys_wait4() from libc.a.
1790 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1792 return sys_wait4(pid, stat_addr, options, NULL);