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>
57 #include "cred-internals.h"
59 static void exit_mm(struct task_struct * tsk);
61 static void __unhash_process(struct task_struct *p)
64 detach_pid(p, PIDTYPE_PID);
65 if (thread_group_leader(p)) {
66 detach_pid(p, PIDTYPE_PGID);
67 detach_pid(p, PIDTYPE_SID);
69 list_del_rcu(&p->tasks);
70 __get_cpu_var(process_counts)--;
72 list_del_rcu(&p->thread_group);
73 list_del_init(&p->sibling);
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 struct sighand_struct *sighand;
85 BUG_ON(!atomic_read(&sig->count));
87 sighand = rcu_dereference(tsk->sighand);
88 spin_lock(&sighand->siglock);
90 posix_cpu_timers_exit(tsk);
91 if (atomic_dec_and_test(&sig->count))
92 posix_cpu_timers_exit_group(tsk);
94 cputime_t utime, stime;
97 * If there is any task waiting for the group exit
100 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
101 wake_up_process(sig->group_exit_task);
103 if (tsk == sig->curr_target)
104 sig->curr_target = next_thread(tsk);
106 * Accumulate here the counters for all threads but the
107 * group leader as they die, so they can be added into
108 * the process-wide totals when those are taken.
109 * The group leader stays around as a zombie as long
110 * as there are other threads. When it gets reaped,
111 * the exit.c code will add its counts into these totals.
112 * We won't ever get here for the group leader, since it
113 * will have been the last reference on the signal_struct.
115 task_times(tsk, &utime, &stime);
116 sig->utime = cputime_add(sig->utime, utime);
117 sig->stime = cputime_add(sig->stime, stime);
118 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
119 sig->min_flt += tsk->min_flt;
120 sig->maj_flt += tsk->maj_flt;
121 sig->nvcsw += tsk->nvcsw;
122 sig->nivcsw += tsk->nivcsw;
123 sig->inblock += task_io_get_inblock(tsk);
124 sig->oublock += task_io_get_oublock(tsk);
125 task_io_accounting_add(&sig->ioac, &tsk->ioac);
126 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
127 sig = NULL; /* Marker for below. */
130 __unhash_process(tsk);
133 * Do this under ->siglock, we can race with another thread
134 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
136 flush_sigqueue(&tsk->pending);
140 spin_unlock(&sighand->siglock);
142 __cleanup_sighand(sighand);
143 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
145 flush_sigqueue(&sig->shared_pending);
146 taskstats_tgid_free(sig);
148 * Make sure ->signal can't go away under rq->lock,
149 * see account_group_exec_runtime().
151 task_rq_unlock_wait(tsk);
152 __cleanup_signal(sig);
156 static void delayed_put_task_struct(struct rcu_head *rhp)
158 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
160 #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 */
176 atomic_dec(&__task_cred(p)->user->processes);
180 write_lock_irq(&tasklist_lock);
181 tracehook_finish_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));
193 do_notify_parent(leader, leader->exit_signal);
195 * If we were the last child thread and the leader has
196 * exited already, and the leader's parent ignores SIGCHLD,
197 * then we are the one who should release the leader.
199 * do_notify_parent() will have marked it self-reaping in
202 zap_leader = task_detached(leader);
205 * This maintains the invariant that release_task()
206 * only runs on a task in EXIT_DEAD, just for sanity.
209 leader->exit_state = EXIT_DEAD;
212 write_unlock_irq(&tasklist_lock);
214 call_rcu(&p->rcu, delayed_put_task_struct);
217 if (unlikely(zap_leader))
222 * This checks not only the pgrp, but falls back on the pid if no
223 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
226 * The caller must hold rcu lock or the tasklist lock.
228 struct pid *session_of_pgrp(struct pid *pgrp)
230 struct task_struct *p;
231 struct pid *sid = NULL;
233 p = pid_task(pgrp, PIDTYPE_PGID);
235 p = pid_task(pgrp, PIDTYPE_PID);
237 sid = task_session(p);
243 * Determine if a process group is "orphaned", according to the POSIX
244 * definition in 2.2.2.52. Orphaned process groups are not to be affected
245 * by terminal-generated stop signals. Newly orphaned process groups are
246 * to receive a SIGHUP and a SIGCONT.
248 * "I ask you, have you ever known what it is to be an orphan?"
250 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
252 struct task_struct *p;
254 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
255 if ((p == ignored_task) ||
256 (p->exit_state && thread_group_empty(p)) ||
257 is_global_init(p->real_parent))
260 if (task_pgrp(p->real_parent) != pgrp &&
261 task_session(p->real_parent) == task_session(p))
263 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
268 int is_current_pgrp_orphaned(void)
272 read_lock(&tasklist_lock);
273 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
274 read_unlock(&tasklist_lock);
279 static int has_stopped_jobs(struct pid *pgrp)
282 struct task_struct *p;
284 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
285 if (!task_is_stopped(p))
289 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
294 * Check to see if any process groups have become orphaned as
295 * a result of our exiting, and if they have any stopped jobs,
296 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
299 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
301 struct pid *pgrp = task_pgrp(tsk);
302 struct task_struct *ignored_task = tsk;
305 /* exit: our father is in a different pgrp than
306 * we are and we were the only connection outside.
308 parent = tsk->real_parent;
310 /* reparent: our child is in a different pgrp than
311 * we are, and it was the only connection outside.
315 if (task_pgrp(parent) != pgrp &&
316 task_session(parent) == task_session(tsk) &&
317 will_become_orphaned_pgrp(pgrp, ignored_task) &&
318 has_stopped_jobs(pgrp)) {
319 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
320 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
325 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
327 * If a kernel thread is launched as a result of a system call, or if
328 * it ever exits, it should generally reparent itself to kthreadd so it
329 * isn't in the way of other processes and is correctly cleaned up on exit.
331 * The various task state such as scheduling policy and priority may have
332 * been inherited from a user process, so we reset them to sane values here.
334 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
336 static void reparent_to_kthreadd(void)
338 write_lock_irq(&tasklist_lock);
340 ptrace_unlink(current);
341 /* Reparent to init */
342 current->real_parent = current->parent = kthreadd_task;
343 list_move_tail(¤t->sibling, ¤t->real_parent->children);
345 /* Set the exit signal to SIGCHLD so we signal init on exit */
346 current->exit_signal = SIGCHLD;
348 if (task_nice(current) < 0)
349 set_user_nice(current, 0);
353 memcpy(current->signal->rlim, init_task.signal->rlim,
354 sizeof(current->signal->rlim));
356 atomic_inc(&init_cred.usage);
357 commit_creds(&init_cred);
358 write_unlock_irq(&tasklist_lock);
361 void __set_special_pids(struct pid *pid)
363 struct task_struct *curr = current->group_leader;
365 if (task_session(curr) != pid)
366 change_pid(curr, PIDTYPE_SID, pid);
368 if (task_pgrp(curr) != pid)
369 change_pid(curr, PIDTYPE_PGID, pid);
372 static void set_special_pids(struct pid *pid)
374 write_lock_irq(&tasklist_lock);
375 __set_special_pids(pid);
376 write_unlock_irq(&tasklist_lock);
380 * Let kernel threads use this to say that they allow a certain signal.
381 * Must not be used if kthread was cloned with CLONE_SIGHAND.
383 int allow_signal(int sig)
385 if (!valid_signal(sig) || sig < 1)
388 spin_lock_irq(¤t->sighand->siglock);
389 /* This is only needed for daemonize()'ed kthreads */
390 sigdelset(¤t->blocked, sig);
392 * Kernel threads handle their own signals. Let the signal code
393 * know it'll be handled, so that they don't get converted to
394 * SIGKILL or just silently dropped.
396 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
398 spin_unlock_irq(¤t->sighand->siglock);
402 EXPORT_SYMBOL(allow_signal);
404 int disallow_signal(int sig)
406 if (!valid_signal(sig) || sig < 1)
409 spin_lock_irq(¤t->sighand->siglock);
410 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
412 spin_unlock_irq(¤t->sighand->siglock);
416 EXPORT_SYMBOL(disallow_signal);
419 * Put all the gunge required to become a kernel thread without
420 * attached user resources in one place where it belongs.
423 void daemonize(const char *name, ...)
428 va_start(args, name);
429 vsnprintf(current->comm, sizeof(current->comm), name, args);
433 * If we were started as result of loading a module, close all of the
434 * user space pages. We don't need them, and if we didn't close them
435 * they would be locked into memory.
439 * We don't want to have TIF_FREEZE set if the system-wide hibernation
440 * or suspend transition begins right now.
442 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
444 if (current->nsproxy != &init_nsproxy) {
445 get_nsproxy(&init_nsproxy);
446 switch_task_namespaces(current, &init_nsproxy);
448 set_special_pids(&init_struct_pid);
449 proc_clear_tty(current);
451 /* Block and flush all signals */
452 sigfillset(&blocked);
453 sigprocmask(SIG_BLOCK, &blocked, NULL);
454 flush_signals(current);
456 /* Become as one with the init task */
458 daemonize_fs_struct();
460 current->files = init_task.files;
461 atomic_inc(¤t->files->count);
463 reparent_to_kthreadd();
466 EXPORT_SYMBOL(daemonize);
468 static void close_files(struct files_struct * files)
476 * It is safe to dereference the fd table without RCU or
477 * ->file_lock because this is the last reference to the
480 fdt = files_fdtable(files);
484 if (i >= fdt->max_fds)
486 set = fdt->open_fds->fds_bits[j++];
489 struct file * file = xchg(&fdt->fd[i], NULL);
491 filp_close(file, files);
501 struct files_struct *get_files_struct(struct task_struct *task)
503 struct files_struct *files;
508 atomic_inc(&files->count);
514 void put_files_struct(struct files_struct *files)
518 if (atomic_dec_and_test(&files->count)) {
521 * Free the fd and fdset arrays if we expanded them.
522 * If the fdtable was embedded, pass files for freeing
523 * at the end of the RCU grace period. Otherwise,
524 * you can free files immediately.
526 fdt = files_fdtable(files);
527 if (fdt != &files->fdtab)
528 kmem_cache_free(files_cachep, files);
533 void reset_files_struct(struct files_struct *files)
535 struct task_struct *tsk = current;
536 struct files_struct *old;
542 put_files_struct(old);
545 void exit_files(struct task_struct *tsk)
547 struct files_struct * files = tsk->files;
553 put_files_struct(files);
557 #ifdef CONFIG_MM_OWNER
559 * Task p is exiting and it owned mm, lets find a new owner for it
562 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
565 * If there are other users of the mm and the owner (us) is exiting
566 * we need to find a new owner to take on the responsibility.
568 if (atomic_read(&mm->mm_users) <= 1)
575 void mm_update_next_owner(struct mm_struct *mm)
577 struct task_struct *c, *g, *p = current;
580 if (!mm_need_new_owner(mm, p))
583 read_lock(&tasklist_lock);
585 * Search in the children
587 list_for_each_entry(c, &p->children, sibling) {
589 goto assign_new_owner;
593 * Search in the siblings
595 list_for_each_entry(c, &p->real_parent->children, sibling) {
597 goto assign_new_owner;
601 * Search through everything else. We should not get
604 do_each_thread(g, c) {
606 goto assign_new_owner;
607 } while_each_thread(g, c);
609 read_unlock(&tasklist_lock);
611 * We found no owner yet mm_users > 1: this implies that we are
612 * most likely racing with swapoff (try_to_unuse()) or /proc or
613 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
622 * The task_lock protects c->mm from changing.
623 * We always want mm->owner->mm == mm
627 * Delay read_unlock() till we have the task_lock()
628 * to ensure that c does not slip away underneath us
630 read_unlock(&tasklist_lock);
640 #endif /* CONFIG_MM_OWNER */
643 * Turn us into a lazy TLB process if we
646 static void exit_mm(struct task_struct * tsk)
648 struct mm_struct *mm = tsk->mm;
649 struct core_state *core_state;
655 * Serialize with any possible pending coredump.
656 * We must hold mmap_sem around checking core_state
657 * and clearing tsk->mm. The core-inducing thread
658 * will increment ->nr_threads for each thread in the
659 * group with ->mm != NULL.
661 down_read(&mm->mmap_sem);
662 core_state = mm->core_state;
664 struct core_thread self;
665 up_read(&mm->mmap_sem);
668 self.next = xchg(&core_state->dumper.next, &self);
670 * Implies mb(), the result of xchg() must be visible
671 * to core_state->dumper.
673 if (atomic_dec_and_test(&core_state->nr_threads))
674 complete(&core_state->startup);
677 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
678 if (!self.task) /* see coredump_finish() */
682 __set_task_state(tsk, TASK_RUNNING);
683 down_read(&mm->mmap_sem);
685 atomic_inc(&mm->mm_count);
686 BUG_ON(mm != tsk->active_mm);
687 /* more a memory barrier than a real lock */
690 up_read(&mm->mmap_sem);
691 enter_lazy_tlb(mm, current);
692 /* We don't want this task to be frozen prematurely */
693 clear_freeze_flag(tsk);
695 mm_update_next_owner(mm);
700 * When we die, we re-parent all our children.
701 * Try to give them to another thread in our thread
702 * group, and if no such member exists, give it to
703 * the child reaper process (ie "init") in our pid
706 static struct task_struct *find_new_reaper(struct task_struct *father)
708 struct pid_namespace *pid_ns = task_active_pid_ns(father);
709 struct task_struct *thread;
712 while_each_thread(father, thread) {
713 if (thread->flags & PF_EXITING)
715 if (unlikely(pid_ns->child_reaper == father))
716 pid_ns->child_reaper = thread;
720 if (unlikely(pid_ns->child_reaper == father)) {
721 write_unlock_irq(&tasklist_lock);
722 if (unlikely(pid_ns == &init_pid_ns))
723 panic("Attempted to kill init!");
725 zap_pid_ns_processes(pid_ns);
726 write_lock_irq(&tasklist_lock);
728 * We can not clear ->child_reaper or leave it alone.
729 * There may by stealth EXIT_DEAD tasks on ->children,
730 * forget_original_parent() must move them somewhere.
732 pid_ns->child_reaper = init_pid_ns.child_reaper;
735 return pid_ns->child_reaper;
739 * Any that need to be release_task'd are put on the @dead list.
741 static void reparent_thread(struct task_struct *father, struct task_struct *p,
742 struct list_head *dead)
744 if (p->pdeath_signal)
745 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
747 list_move_tail(&p->sibling, &p->real_parent->children);
749 if (task_detached(p))
752 * If this is a threaded reparent there is no need to
753 * notify anyone anything has happened.
755 if (same_thread_group(p->real_parent, father))
758 /* We don't want people slaying init. */
759 p->exit_signal = SIGCHLD;
761 /* If it has exited notify the new parent about this child's death. */
762 if (!task_ptrace(p) &&
763 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
764 do_notify_parent(p, p->exit_signal);
765 if (task_detached(p)) {
766 p->exit_state = EXIT_DEAD;
767 list_move_tail(&p->sibling, dead);
771 kill_orphaned_pgrp(p, father);
774 static void forget_original_parent(struct task_struct *father)
776 struct task_struct *p, *n, *reaper;
777 LIST_HEAD(dead_children);
781 write_lock_irq(&tasklist_lock);
782 reaper = find_new_reaper(father);
784 list_for_each_entry_safe(p, n, &father->children, sibling) {
785 p->real_parent = reaper;
786 if (p->parent == father) {
787 BUG_ON(task_ptrace(p));
788 p->parent = p->real_parent;
790 reparent_thread(father, p, &dead_children);
792 write_unlock_irq(&tasklist_lock);
794 BUG_ON(!list_empty(&father->children));
796 list_for_each_entry_safe(p, n, &dead_children, sibling) {
797 list_del_init(&p->sibling);
803 * Send signals to all our closest relatives so that they know
804 * to properly mourn us..
806 static void exit_notify(struct task_struct *tsk, int group_dead)
812 * This does two things:
814 * A. Make init inherit all the child processes
815 * B. Check to see if any process groups have become orphaned
816 * as a result of our exiting, and if they have any stopped
817 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
819 forget_original_parent(tsk);
820 exit_task_namespaces(tsk);
822 write_lock_irq(&tasklist_lock);
824 kill_orphaned_pgrp(tsk->group_leader, NULL);
826 /* Let father know we died
828 * Thread signals are configurable, but you aren't going to use
829 * that to send signals to arbitary processes.
830 * That stops right now.
832 * If the parent exec id doesn't match the exec id we saved
833 * when we started then we know the parent has changed security
836 * If our self_exec id doesn't match our parent_exec_id then
837 * we have changed execution domain as these two values started
838 * the same after a fork.
840 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
841 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
842 tsk->self_exec_id != tsk->parent_exec_id))
843 tsk->exit_signal = SIGCHLD;
845 signal = tracehook_notify_death(tsk, &cookie, group_dead);
847 signal = do_notify_parent(tsk, signal);
849 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
851 /* mt-exec, de_thread() is waiting for us */
852 if (thread_group_leader(tsk) &&
853 tsk->signal->group_exit_task &&
854 tsk->signal->notify_count < 0)
855 wake_up_process(tsk->signal->group_exit_task);
857 write_unlock_irq(&tasklist_lock);
859 tracehook_report_death(tsk, signal, cookie, group_dead);
861 /* If the process is dead, release it - nobody will wait for it */
862 if (signal == DEATH_REAP)
866 #ifdef CONFIG_DEBUG_STACK_USAGE
867 static void check_stack_usage(void)
869 static DEFINE_SPINLOCK(low_water_lock);
870 static int lowest_to_date = THREAD_SIZE;
873 free = stack_not_used(current);
875 if (free >= lowest_to_date)
878 spin_lock(&low_water_lock);
879 if (free < lowest_to_date) {
880 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
882 current->comm, free);
883 lowest_to_date = free;
885 spin_unlock(&low_water_lock);
888 static inline void check_stack_usage(void) {}
891 NORET_TYPE void do_exit(long code)
893 struct task_struct *tsk = current;
896 profile_task_exit(tsk);
898 WARN_ON(atomic_read(&tsk->fs_excl));
900 if (unlikely(in_interrupt()))
901 panic("Aiee, killing interrupt handler!");
902 if (unlikely(!tsk->pid))
903 panic("Attempted to kill the idle task!");
905 tracehook_report_exit(&code);
907 validate_creds_for_do_exit(tsk);
910 * We're taking recursive faults here in do_exit. Safest is to just
911 * leave this task alone and wait for reboot.
913 if (unlikely(tsk->flags & PF_EXITING)) {
915 "Fixing recursive fault but reboot is needed!\n");
917 * We can do this unlocked here. The futex code uses
918 * this flag just to verify whether the pi state
919 * cleanup has been done or not. In the worst case it
920 * loops once more. We pretend that the cleanup was
921 * done as there is no way to return. Either the
922 * OWNER_DIED bit is set by now or we push the blocked
923 * task into the wait for ever nirwana as well.
925 tsk->flags |= PF_EXITPIDONE;
926 set_current_state(TASK_UNINTERRUPTIBLE);
932 exit_signals(tsk); /* sets PF_EXITING */
934 * tsk->flags are checked in the futex code to protect against
935 * an exiting task cleaning up the robust pi futexes.
938 spin_unlock_wait(&tsk->pi_lock);
940 if (unlikely(in_atomic()))
941 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
942 current->comm, task_pid_nr(current),
945 acct_update_integrals(tsk);
947 group_dead = atomic_dec_and_test(&tsk->signal->live);
949 hrtimer_cancel(&tsk->signal->real_timer);
950 exit_itimers(tsk->signal);
952 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
954 acct_collect(code, group_dead);
957 if (unlikely(tsk->audit_context))
960 tsk->exit_code = code;
961 taskstats_exit(tsk, group_dead);
967 trace_sched_process_exit(tsk);
976 if (group_dead && tsk->signal->leader)
977 disassociate_ctty(1);
979 module_put(task_thread_info(tsk)->exec_domain->module);
981 proc_exit_connector(tsk);
984 * Flush inherited counters to the parent - before the parent
985 * gets woken up by child-exit notifications.
987 perf_event_exit_task(tsk);
989 exit_notify(tsk, group_dead);
991 mpol_put(tsk->mempolicy);
992 tsk->mempolicy = NULL;
995 if (unlikely(current->pi_state_cache))
996 kfree(current->pi_state_cache);
999 * Make sure we are holding no locks:
1001 debug_check_no_locks_held(tsk);
1003 * We can do this unlocked here. The futex code uses this flag
1004 * just to verify whether the pi state cleanup has been done
1005 * or not. In the worst case it loops once more.
1007 tsk->flags |= PF_EXITPIDONE;
1009 if (tsk->io_context)
1012 if (tsk->splice_pipe)
1013 __free_pipe_info(tsk->splice_pipe);
1015 validate_creds_for_do_exit(tsk);
1019 /* causes final put_task_struct in finish_task_switch(). */
1020 tsk->state = TASK_DEAD;
1023 /* Avoid "noreturn function does return". */
1025 cpu_relax(); /* For when BUG is null */
1028 EXPORT_SYMBOL_GPL(do_exit);
1030 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1038 EXPORT_SYMBOL(complete_and_exit);
1040 SYSCALL_DEFINE1(exit, int, error_code)
1042 do_exit((error_code&0xff)<<8);
1046 * Take down every thread in the group. This is called by fatal signals
1047 * as well as by sys_exit_group (below).
1050 do_group_exit(int exit_code)
1052 struct signal_struct *sig = current->signal;
1054 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1056 if (signal_group_exit(sig))
1057 exit_code = sig->group_exit_code;
1058 else if (!thread_group_empty(current)) {
1059 struct sighand_struct *const sighand = current->sighand;
1060 spin_lock_irq(&sighand->siglock);
1061 if (signal_group_exit(sig))
1062 /* Another thread got here before we took the lock. */
1063 exit_code = sig->group_exit_code;
1065 sig->group_exit_code = exit_code;
1066 sig->flags = SIGNAL_GROUP_EXIT;
1067 zap_other_threads(current);
1069 spin_unlock_irq(&sighand->siglock);
1077 * this kills every thread in the thread group. Note that any externally
1078 * wait4()-ing process will get the correct exit code - even if this
1079 * thread is not the thread group leader.
1081 SYSCALL_DEFINE1(exit_group, int, error_code)
1083 do_group_exit((error_code & 0xff) << 8);
1089 enum pid_type wo_type;
1093 struct siginfo __user *wo_info;
1094 int __user *wo_stat;
1095 struct rusage __user *wo_rusage;
1097 wait_queue_t child_wait;
1102 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1104 if (type != PIDTYPE_PID)
1105 task = task->group_leader;
1106 return task->pids[type].pid;
1109 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1111 return wo->wo_type == PIDTYPE_MAX ||
1112 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1115 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1117 if (!eligible_pid(wo, p))
1119 /* Wait for all children (clone and not) if __WALL is set;
1120 * otherwise, wait for clone children *only* if __WCLONE is
1121 * set; otherwise, wait for non-clone children *only*. (Note:
1122 * A "clone" child here is one that reports to its parent
1123 * using a signal other than SIGCHLD.) */
1124 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1125 && !(wo->wo_flags & __WALL))
1131 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1132 pid_t pid, uid_t uid, int why, int status)
1134 struct siginfo __user *infop;
1135 int retval = wo->wo_rusage
1136 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1139 infop = wo->wo_info;
1142 retval = put_user(SIGCHLD, &infop->si_signo);
1144 retval = put_user(0, &infop->si_errno);
1146 retval = put_user((short)why, &infop->si_code);
1148 retval = put_user(pid, &infop->si_pid);
1150 retval = put_user(uid, &infop->si_uid);
1152 retval = put_user(status, &infop->si_status);
1160 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1161 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1162 * the lock and this task is uninteresting. If we return nonzero, we have
1163 * released the lock and the system call should return.
1165 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1167 unsigned long state;
1168 int retval, status, traced;
1169 pid_t pid = task_pid_vnr(p);
1170 uid_t uid = __task_cred(p)->uid;
1171 struct siginfo __user *infop;
1173 if (!likely(wo->wo_flags & WEXITED))
1176 if (unlikely(wo->wo_flags & WNOWAIT)) {
1177 int exit_code = p->exit_code;
1181 read_unlock(&tasklist_lock);
1182 if ((exit_code & 0x7f) == 0) {
1184 status = exit_code >> 8;
1186 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1187 status = exit_code & 0x7f;
1189 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1193 * Try to move the task's state to DEAD
1194 * only one thread is allowed to do this:
1196 state = xchg(&p->exit_state, EXIT_DEAD);
1197 if (state != EXIT_ZOMBIE) {
1198 BUG_ON(state != EXIT_DEAD);
1202 traced = ptrace_reparented(p);
1204 * It can be ptraced but not reparented, check
1205 * !task_detached() to filter out sub-threads.
1207 if (likely(!traced) && likely(!task_detached(p))) {
1208 struct signal_struct *psig;
1209 struct signal_struct *sig;
1210 unsigned long maxrss;
1213 * The resource counters for the group leader are in its
1214 * own task_struct. Those for dead threads in the group
1215 * are in its signal_struct, as are those for the child
1216 * processes it has previously reaped. All these
1217 * accumulate in the parent's signal_struct c* fields.
1219 * We don't bother to take a lock here to protect these
1220 * p->signal fields, because they are only touched by
1221 * __exit_signal, which runs with tasklist_lock
1222 * write-locked anyway, and so is excluded here. We do
1223 * need to protect the access to parent->signal fields,
1224 * as other threads in the parent group can be right
1225 * here reaping other children at the same time.
1227 spin_lock_irq(&p->real_parent->sighand->siglock);
1228 psig = p->real_parent->signal;
1231 cputime_add(psig->cutime,
1232 cputime_add(p->utime,
1233 cputime_add(sig->utime,
1236 cputime_add(psig->cstime,
1237 cputime_add(p->stime,
1238 cputime_add(sig->stime,
1241 cputime_add(psig->cgtime,
1242 cputime_add(p->gtime,
1243 cputime_add(sig->gtime,
1246 p->min_flt + sig->min_flt + sig->cmin_flt;
1248 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1250 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1252 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1254 task_io_get_inblock(p) +
1255 sig->inblock + sig->cinblock;
1257 task_io_get_oublock(p) +
1258 sig->oublock + sig->coublock;
1259 maxrss = max(sig->maxrss, sig->cmaxrss);
1260 if (psig->cmaxrss < maxrss)
1261 psig->cmaxrss = maxrss;
1262 task_io_accounting_add(&psig->ioac, &p->ioac);
1263 task_io_accounting_add(&psig->ioac, &sig->ioac);
1264 spin_unlock_irq(&p->real_parent->sighand->siglock);
1268 * Now we are sure this task is interesting, and no other
1269 * thread can reap it because we set its state to EXIT_DEAD.
1271 read_unlock(&tasklist_lock);
1273 retval = wo->wo_rusage
1274 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1275 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1276 ? p->signal->group_exit_code : p->exit_code;
1277 if (!retval && wo->wo_stat)
1278 retval = put_user(status, wo->wo_stat);
1280 infop = wo->wo_info;
1281 if (!retval && infop)
1282 retval = put_user(SIGCHLD, &infop->si_signo);
1283 if (!retval && infop)
1284 retval = put_user(0, &infop->si_errno);
1285 if (!retval && infop) {
1288 if ((status & 0x7f) == 0) {
1292 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1295 retval = put_user((short)why, &infop->si_code);
1297 retval = put_user(status, &infop->si_status);
1299 if (!retval && infop)
1300 retval = put_user(pid, &infop->si_pid);
1301 if (!retval && infop)
1302 retval = put_user(uid, &infop->si_uid);
1307 write_lock_irq(&tasklist_lock);
1308 /* We dropped tasklist, ptracer could die and untrace */
1311 * If this is not a detached task, notify the parent.
1312 * If it's still not detached after that, don't release
1315 if (!task_detached(p)) {
1316 do_notify_parent(p, p->exit_signal);
1317 if (!task_detached(p)) {
1318 p->exit_state = EXIT_ZOMBIE;
1322 write_unlock_irq(&tasklist_lock);
1330 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1333 if (task_is_stopped_or_traced(p))
1334 return &p->exit_code;
1336 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1337 return &p->signal->group_exit_code;
1343 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1344 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1345 * the lock and this task is uninteresting. If we return nonzero, we have
1346 * released the lock and the system call should return.
1348 static int wait_task_stopped(struct wait_opts *wo,
1349 int ptrace, struct task_struct *p)
1351 struct siginfo __user *infop;
1352 int retval, exit_code, *p_code, why;
1353 uid_t uid = 0; /* unneeded, required by compiler */
1357 * Traditionally we see ptrace'd stopped tasks regardless of options.
1359 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1363 spin_lock_irq(&p->sighand->siglock);
1365 p_code = task_stopped_code(p, ptrace);
1366 if (unlikely(!p_code))
1369 exit_code = *p_code;
1373 if (!unlikely(wo->wo_flags & WNOWAIT))
1376 /* don't need the RCU readlock here as we're holding a spinlock */
1377 uid = __task_cred(p)->uid;
1379 spin_unlock_irq(&p->sighand->siglock);
1384 * Now we are pretty sure this task is interesting.
1385 * Make sure it doesn't get reaped out from under us while we
1386 * give up the lock and then examine it below. We don't want to
1387 * keep holding onto the tasklist_lock while we call getrusage and
1388 * possibly take page faults for user memory.
1391 pid = task_pid_vnr(p);
1392 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1393 read_unlock(&tasklist_lock);
1395 if (unlikely(wo->wo_flags & WNOWAIT))
1396 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1398 retval = wo->wo_rusage
1399 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1400 if (!retval && wo->wo_stat)
1401 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1403 infop = wo->wo_info;
1404 if (!retval && infop)
1405 retval = put_user(SIGCHLD, &infop->si_signo);
1406 if (!retval && infop)
1407 retval = put_user(0, &infop->si_errno);
1408 if (!retval && infop)
1409 retval = put_user((short)why, &infop->si_code);
1410 if (!retval && infop)
1411 retval = put_user(exit_code, &infop->si_status);
1412 if (!retval && infop)
1413 retval = put_user(pid, &infop->si_pid);
1414 if (!retval && infop)
1415 retval = put_user(uid, &infop->si_uid);
1425 * Handle do_wait work for one task in a live, non-stopped state.
1426 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1427 * the lock and this task is uninteresting. If we return nonzero, we have
1428 * released the lock and the system call should return.
1430 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1436 if (!unlikely(wo->wo_flags & WCONTINUED))
1439 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1442 spin_lock_irq(&p->sighand->siglock);
1443 /* Re-check with the lock held. */
1444 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1445 spin_unlock_irq(&p->sighand->siglock);
1448 if (!unlikely(wo->wo_flags & WNOWAIT))
1449 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1450 uid = __task_cred(p)->uid;
1451 spin_unlock_irq(&p->sighand->siglock);
1453 pid = task_pid_vnr(p);
1455 read_unlock(&tasklist_lock);
1458 retval = wo->wo_rusage
1459 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1461 if (!retval && wo->wo_stat)
1462 retval = put_user(0xffff, wo->wo_stat);
1466 retval = wait_noreap_copyout(wo, p, pid, uid,
1467 CLD_CONTINUED, SIGCONT);
1468 BUG_ON(retval == 0);
1475 * Consider @p for a wait by @parent.
1477 * -ECHILD should be in ->notask_error before the first call.
1478 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1479 * Returns zero if the search for a child should continue;
1480 * then ->notask_error is 0 if @p is an eligible child,
1481 * or another error from security_task_wait(), or still -ECHILD.
1483 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1484 struct task_struct *p)
1486 int ret = eligible_child(wo, p);
1490 ret = security_task_wait(p);
1491 if (unlikely(ret < 0)) {
1493 * If we have not yet seen any eligible child,
1494 * then let this error code replace -ECHILD.
1495 * A permission error will give the user a clue
1496 * to look for security policy problems, rather
1497 * than for mysterious wait bugs.
1499 if (wo->notask_error)
1500 wo->notask_error = ret;
1504 if (likely(!ptrace) && unlikely(task_ptrace(p))) {
1506 * This child is hidden by ptrace.
1507 * We aren't allowed to see it now, but eventually we will.
1509 wo->notask_error = 0;
1513 if (p->exit_state == EXIT_DEAD)
1517 * We don't reap group leaders with subthreads.
1519 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1520 return wait_task_zombie(wo, p);
1523 * It's stopped or running now, so it might
1524 * later continue, exit, or stop again.
1526 wo->notask_error = 0;
1528 if (task_stopped_code(p, ptrace))
1529 return wait_task_stopped(wo, ptrace, p);
1531 return wait_task_continued(wo, p);
1535 * Do the work of do_wait() for one thread in the group, @tsk.
1537 * -ECHILD should be in ->notask_error before the first call.
1538 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1539 * Returns zero if the search for a child should continue; then
1540 * ->notask_error is 0 if there were any eligible children,
1541 * or another error from security_task_wait(), or still -ECHILD.
1543 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1545 struct task_struct *p;
1547 list_for_each_entry(p, &tsk->children, sibling) {
1549 * Do not consider detached threads.
1551 if (!task_detached(p)) {
1552 int ret = wait_consider_task(wo, 0, p);
1561 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1563 struct task_struct *p;
1565 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1566 int ret = wait_consider_task(wo, 1, p);
1574 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1575 int sync, void *key)
1577 struct wait_opts *wo = container_of(wait, struct wait_opts,
1579 struct task_struct *p = key;
1581 if (!eligible_pid(wo, p))
1584 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1587 return default_wake_function(wait, mode, sync, key);
1590 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1592 __wake_up_sync_key(&parent->signal->wait_chldexit,
1593 TASK_INTERRUPTIBLE, 1, p);
1596 static long do_wait(struct wait_opts *wo)
1598 struct task_struct *tsk;
1601 trace_sched_process_wait(wo->wo_pid);
1603 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1604 wo->child_wait.private = current;
1605 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1608 * If there is nothing that can match our critiera just get out.
1609 * We will clear ->notask_error to zero if we see any child that
1610 * might later match our criteria, even if we are not able to reap
1613 wo->notask_error = -ECHILD;
1614 if ((wo->wo_type < PIDTYPE_MAX) &&
1615 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1618 set_current_state(TASK_INTERRUPTIBLE);
1619 read_lock(&tasklist_lock);
1622 retval = do_wait_thread(wo, tsk);
1626 retval = ptrace_do_wait(wo, tsk);
1630 if (wo->wo_flags & __WNOTHREAD)
1632 } while_each_thread(current, tsk);
1633 read_unlock(&tasklist_lock);
1636 retval = wo->notask_error;
1637 if (!retval && !(wo->wo_flags & WNOHANG)) {
1638 retval = -ERESTARTSYS;
1639 if (!signal_pending(current)) {
1645 __set_current_state(TASK_RUNNING);
1646 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1650 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1651 infop, int, options, struct rusage __user *, ru)
1653 struct wait_opts wo;
1654 struct pid *pid = NULL;
1658 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1660 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1673 type = PIDTYPE_PGID;
1681 if (type < PIDTYPE_MAX)
1682 pid = find_get_pid(upid);
1686 wo.wo_flags = options;
1696 * For a WNOHANG return, clear out all the fields
1697 * we would set so the user can easily tell the
1701 ret = put_user(0, &infop->si_signo);
1703 ret = put_user(0, &infop->si_errno);
1705 ret = put_user(0, &infop->si_code);
1707 ret = put_user(0, &infop->si_pid);
1709 ret = put_user(0, &infop->si_uid);
1711 ret = put_user(0, &infop->si_status);
1716 /* avoid REGPARM breakage on x86: */
1717 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1721 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1722 int, options, struct rusage __user *, ru)
1724 struct wait_opts wo;
1725 struct pid *pid = NULL;
1729 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1730 __WNOTHREAD|__WCLONE|__WALL))
1735 else if (upid < 0) {
1736 type = PIDTYPE_PGID;
1737 pid = find_get_pid(-upid);
1738 } else if (upid == 0) {
1739 type = PIDTYPE_PGID;
1740 pid = get_task_pid(current, PIDTYPE_PGID);
1741 } else /* upid > 0 */ {
1743 pid = find_get_pid(upid);
1748 wo.wo_flags = options | WEXITED;
1750 wo.wo_stat = stat_addr;
1755 /* avoid REGPARM breakage on x86: */
1756 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1760 #ifdef __ARCH_WANT_SYS_WAITPID
1763 * sys_waitpid() remains for compatibility. waitpid() should be
1764 * implemented by calling sys_wait4() from libc.a.
1766 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1768 return sys_wait4(pid, stat_addr, options, NULL);