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/mnt_namespace.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/binfmts.h>
23 #include <linux/nsproxy.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/ptrace.h>
26 #include <linux/profile.h>
27 #include <linux/mount.h>
28 #include <linux/proc_fs.h>
29 #include <linux/kthread.h>
30 #include <linux/mempolicy.h>
31 #include <linux/taskstats_kern.h>
32 #include <linux/delayacct.h>
33 #include <linux/freezer.h>
34 #include <linux/cgroup.h>
35 #include <linux/syscalls.h>
36 #include <linux/signal.h>
37 #include <linux/posix-timers.h>
38 #include <linux/cn_proc.h>
39 #include <linux/mutex.h>
40 #include <linux/futex.h>
41 #include <linux/compat.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>
48 #include <asm/uaccess.h>
49 #include <asm/unistd.h>
50 #include <asm/pgtable.h>
51 #include <asm/mmu_context.h>
53 static void exit_mm(struct task_struct * tsk);
55 static void __unhash_process(struct task_struct *p)
58 detach_pid(p, PIDTYPE_PID);
59 if (thread_group_leader(p)) {
60 detach_pid(p, PIDTYPE_PGID);
61 detach_pid(p, PIDTYPE_SID);
63 list_del_rcu(&p->tasks);
64 __get_cpu_var(process_counts)--;
66 list_del_rcu(&p->thread_group);
71 * This function expects the tasklist_lock write-locked.
73 static void __exit_signal(struct task_struct *tsk)
75 struct signal_struct *sig = tsk->signal;
76 struct sighand_struct *sighand;
79 BUG_ON(!atomic_read(&sig->count));
82 sighand = rcu_dereference(tsk->sighand);
83 spin_lock(&sighand->siglock);
85 posix_cpu_timers_exit(tsk);
86 if (atomic_dec_and_test(&sig->count))
87 posix_cpu_timers_exit_group(tsk);
90 * If there is any task waiting for the group exit
93 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
94 wake_up_process(sig->group_exit_task);
96 if (tsk == sig->curr_target)
97 sig->curr_target = next_thread(tsk);
99 * Accumulate here the counters for all threads but the
100 * group leader as they die, so they can be added into
101 * the process-wide totals when those are taken.
102 * The group leader stays around as a zombie as long
103 * as there are other threads. When it gets reaped,
104 * the exit.c code will add its counts into these totals.
105 * We won't ever get here for the group leader, since it
106 * will have been the last reference on the signal_struct.
108 sig->utime = cputime_add(sig->utime, tsk->utime);
109 sig->stime = cputime_add(sig->stime, tsk->stime);
110 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
111 sig->min_flt += tsk->min_flt;
112 sig->maj_flt += tsk->maj_flt;
113 sig->nvcsw += tsk->nvcsw;
114 sig->nivcsw += tsk->nivcsw;
115 sig->inblock += task_io_get_inblock(tsk);
116 sig->oublock += task_io_get_oublock(tsk);
117 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
118 sig = NULL; /* Marker for below. */
121 __unhash_process(tsk);
125 spin_unlock(&sighand->siglock);
128 __cleanup_sighand(sighand);
129 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
130 flush_sigqueue(&tsk->pending);
132 flush_sigqueue(&sig->shared_pending);
133 taskstats_tgid_free(sig);
134 __cleanup_signal(sig);
138 static void delayed_put_task_struct(struct rcu_head *rhp)
140 put_task_struct(container_of(rhp, struct task_struct, rcu));
143 void release_task(struct task_struct * p)
145 struct task_struct *leader;
148 atomic_dec(&p->user->processes);
150 write_lock_irq(&tasklist_lock);
152 BUG_ON(!list_empty(&p->ptrace_list) || !list_empty(&p->ptrace_children));
156 * If we are the last non-leader member of the thread
157 * group, and the leader is zombie, then notify the
158 * group leader's parent process. (if it wants notification.)
161 leader = p->group_leader;
162 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
163 BUG_ON(leader->exit_signal == -1);
164 do_notify_parent(leader, leader->exit_signal);
166 * If we were the last child thread and the leader has
167 * exited already, and the leader's parent ignores SIGCHLD,
168 * then we are the one who should release the leader.
170 * do_notify_parent() will have marked it self-reaping in
173 zap_leader = (leader->exit_signal == -1);
176 write_unlock_irq(&tasklist_lock);
178 call_rcu(&p->rcu, delayed_put_task_struct);
181 if (unlikely(zap_leader))
186 * This checks not only the pgrp, but falls back on the pid if no
187 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
190 * The caller must hold rcu lock or the tasklist lock.
192 struct pid *session_of_pgrp(struct pid *pgrp)
194 struct task_struct *p;
195 struct pid *sid = NULL;
197 p = pid_task(pgrp, PIDTYPE_PGID);
199 p = pid_task(pgrp, PIDTYPE_PID);
201 sid = task_session(p);
207 * Determine if a process group is "orphaned", according to the POSIX
208 * definition in 2.2.2.52. Orphaned process groups are not to be affected
209 * by terminal-generated stop signals. Newly orphaned process groups are
210 * to receive a SIGHUP and a SIGCONT.
212 * "I ask you, have you ever known what it is to be an orphan?"
214 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
216 struct task_struct *p;
218 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
219 if ((p == ignored_task) ||
220 (p->exit_state && thread_group_empty(p)) ||
221 is_global_init(p->real_parent))
224 if (task_pgrp(p->real_parent) != pgrp &&
225 task_session(p->real_parent) == task_session(p))
227 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
232 int is_current_pgrp_orphaned(void)
236 read_lock(&tasklist_lock);
237 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
238 read_unlock(&tasklist_lock);
243 static int has_stopped_jobs(struct pid *pgrp)
246 struct task_struct *p;
248 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
249 if (!task_is_stopped(p))
253 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
258 * Check to see if any process groups have become orphaned as
259 * a result of our exiting, and if they have any stopped jobs,
260 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
263 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
265 struct pid *pgrp = task_pgrp(tsk);
266 struct task_struct *ignored_task = tsk;
269 /* exit: our father is in a different pgrp than
270 * we are and we were the only connection outside.
272 parent = tsk->real_parent;
274 /* reparent: our child is in a different pgrp than
275 * we are, and it was the only connection outside.
279 if (task_pgrp(parent) != pgrp &&
280 task_session(parent) == task_session(tsk) &&
281 will_become_orphaned_pgrp(pgrp, ignored_task) &&
282 has_stopped_jobs(pgrp)) {
283 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
284 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
289 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
291 * If a kernel thread is launched as a result of a system call, or if
292 * it ever exits, it should generally reparent itself to kthreadd so it
293 * isn't in the way of other processes and is correctly cleaned up on exit.
295 * The various task state such as scheduling policy and priority may have
296 * been inherited from a user process, so we reset them to sane values here.
298 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
300 static void reparent_to_kthreadd(void)
302 write_lock_irq(&tasklist_lock);
304 ptrace_unlink(current);
305 /* Reparent to init */
306 remove_parent(current);
307 current->real_parent = current->parent = kthreadd_task;
310 /* Set the exit signal to SIGCHLD so we signal init on exit */
311 current->exit_signal = SIGCHLD;
313 if (task_nice(current) < 0)
314 set_user_nice(current, 0);
318 security_task_reparent_to_init(current);
319 memcpy(current->signal->rlim, init_task.signal->rlim,
320 sizeof(current->signal->rlim));
321 atomic_inc(&(INIT_USER->__count));
322 write_unlock_irq(&tasklist_lock);
323 switch_uid(INIT_USER);
326 void __set_special_pids(struct pid *pid)
328 struct task_struct *curr = current->group_leader;
329 pid_t nr = pid_nr(pid);
331 if (task_session(curr) != pid) {
332 detach_pid(curr, PIDTYPE_SID);
333 attach_pid(curr, PIDTYPE_SID, pid);
334 set_task_session(curr, nr);
336 if (task_pgrp(curr) != pid) {
337 detach_pid(curr, PIDTYPE_PGID);
338 attach_pid(curr, PIDTYPE_PGID, pid);
339 set_task_pgrp(curr, nr);
343 static void set_special_pids(struct pid *pid)
345 write_lock_irq(&tasklist_lock);
346 __set_special_pids(pid);
347 write_unlock_irq(&tasklist_lock);
351 * Let kernel threads use this to say that they
352 * allow a certain signal (since daemonize() will
353 * have disabled all of them by default).
355 int allow_signal(int sig)
357 if (!valid_signal(sig) || sig < 1)
360 spin_lock_irq(¤t->sighand->siglock);
361 sigdelset(¤t->blocked, sig);
363 /* Kernel threads handle their own signals.
364 Let the signal code know it'll be handled, so
365 that they don't get converted to SIGKILL or
366 just silently dropped */
367 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
370 spin_unlock_irq(¤t->sighand->siglock);
374 EXPORT_SYMBOL(allow_signal);
376 int disallow_signal(int sig)
378 if (!valid_signal(sig) || sig < 1)
381 spin_lock_irq(¤t->sighand->siglock);
382 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
384 spin_unlock_irq(¤t->sighand->siglock);
388 EXPORT_SYMBOL(disallow_signal);
391 * Put all the gunge required to become a kernel thread without
392 * attached user resources in one place where it belongs.
395 void daemonize(const char *name, ...)
398 struct fs_struct *fs;
401 va_start(args, name);
402 vsnprintf(current->comm, sizeof(current->comm), name, args);
406 * If we were started as result of loading a module, close all of the
407 * user space pages. We don't need them, and if we didn't close them
408 * they would be locked into memory.
412 * We don't want to have TIF_FREEZE set if the system-wide hibernation
413 * or suspend transition begins right now.
415 current->flags |= PF_NOFREEZE;
417 if (current->nsproxy != &init_nsproxy) {
418 get_nsproxy(&init_nsproxy);
419 switch_task_namespaces(current, &init_nsproxy);
421 set_special_pids(&init_struct_pid);
422 proc_clear_tty(current);
424 /* Block and flush all signals */
425 sigfillset(&blocked);
426 sigprocmask(SIG_BLOCK, &blocked, NULL);
427 flush_signals(current);
429 /* Become as one with the init task */
431 exit_fs(current); /* current->fs->count--; */
434 atomic_inc(&fs->count);
437 current->files = init_task.files;
438 atomic_inc(¤t->files->count);
440 reparent_to_kthreadd();
443 EXPORT_SYMBOL(daemonize);
445 static void close_files(struct files_struct * files)
453 * It is safe to dereference the fd table without RCU or
454 * ->file_lock because this is the last reference to the
457 fdt = files_fdtable(files);
461 if (i >= fdt->max_fds)
463 set = fdt->open_fds->fds_bits[j++];
466 struct file * file = xchg(&fdt->fd[i], NULL);
468 filp_close(file, files);
478 struct files_struct *get_files_struct(struct task_struct *task)
480 struct files_struct *files;
485 atomic_inc(&files->count);
491 void put_files_struct(struct files_struct *files)
495 if (atomic_dec_and_test(&files->count)) {
498 * Free the fd and fdset arrays if we expanded them.
499 * If the fdtable was embedded, pass files for freeing
500 * at the end of the RCU grace period. Otherwise,
501 * you can free files immediately.
503 fdt = files_fdtable(files);
504 if (fdt != &files->fdtab)
505 kmem_cache_free(files_cachep, files);
510 void reset_files_struct(struct files_struct *files)
512 struct task_struct *tsk = current;
513 struct files_struct *old;
519 put_files_struct(old);
522 void exit_files(struct task_struct *tsk)
524 struct files_struct * files = tsk->files;
530 put_files_struct(files);
534 void put_fs_struct(struct fs_struct *fs)
536 /* No need to hold fs->lock if we are killing it */
537 if (atomic_dec_and_test(&fs->count)) {
540 if (fs->altroot.dentry)
541 path_put(&fs->altroot);
542 kmem_cache_free(fs_cachep, fs);
546 void exit_fs(struct task_struct *tsk)
548 struct fs_struct * fs = tsk->fs;
558 EXPORT_SYMBOL_GPL(exit_fs);
560 #ifdef CONFIG_MM_OWNER
562 * Task p is exiting and it owned mm, lets find a new owner for it
565 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
568 * If there are other users of the mm and the owner (us) is exiting
569 * we need to find a new owner to take on the responsibility.
573 if (atomic_read(&mm->mm_users) <= 1)
580 void mm_update_next_owner(struct mm_struct *mm)
582 struct task_struct *c, *g, *p = current;
585 if (!mm_need_new_owner(mm, p))
588 read_lock(&tasklist_lock);
590 * Search in the children
592 list_for_each_entry(c, &p->children, sibling) {
594 goto assign_new_owner;
598 * Search in the siblings
600 list_for_each_entry(c, &p->parent->children, sibling) {
602 goto assign_new_owner;
606 * Search through everything else. We should not get
609 do_each_thread(g, c) {
611 goto assign_new_owner;
612 } while_each_thread(g, c);
614 read_unlock(&tasklist_lock);
621 * The task_lock protects c->mm from changing.
622 * We always want mm->owner->mm == mm
626 * Delay read_unlock() till we have the task_lock()
627 * to ensure that c does not slip away underneath us
629 read_unlock(&tasklist_lock);
635 cgroup_mm_owner_callbacks(mm->owner, c);
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;
654 * Serialize with any possible pending coredump.
655 * We must hold mmap_sem around checking core_waiters
656 * and clearing tsk->mm. The core-inducing thread
657 * will increment core_waiters for each thread in the
658 * group with ->mm != NULL.
660 down_read(&mm->mmap_sem);
661 if (mm->core_waiters) {
662 up_read(&mm->mmap_sem);
663 down_write(&mm->mmap_sem);
664 if (!--mm->core_waiters)
665 complete(mm->core_startup_done);
666 up_write(&mm->mmap_sem);
668 wait_for_completion(&mm->core_done);
669 down_read(&mm->mmap_sem);
671 atomic_inc(&mm->mm_count);
672 BUG_ON(mm != tsk->active_mm);
673 /* more a memory barrier than a real lock */
676 up_read(&mm->mmap_sem);
677 enter_lazy_tlb(mm, current);
678 /* We don't want this task to be frozen prematurely */
679 clear_freeze_flag(tsk);
681 mm_update_next_owner(mm);
686 reparent_thread(struct task_struct *p, struct task_struct *father, int traced)
688 if (p->pdeath_signal)
689 /* We already hold the tasklist_lock here. */
690 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
692 /* Move the child from its dying parent to the new one. */
693 if (unlikely(traced)) {
694 /* Preserve ptrace links if someone else is tracing this child. */
695 list_del_init(&p->ptrace_list);
696 if (p->parent != p->real_parent)
697 list_add(&p->ptrace_list, &p->real_parent->ptrace_children);
699 /* If this child is being traced, then we're the one tracing it
700 * anyway, so let go of it.
704 p->parent = p->real_parent;
707 if (task_is_traced(p)) {
709 * If it was at a trace stop, turn it into
710 * a normal stop since it's no longer being
717 /* If this is a threaded reparent there is no need to
718 * notify anyone anything has happened.
720 if (p->real_parent->group_leader == father->group_leader)
723 /* We don't want people slaying init. */
724 if (p->exit_signal != -1)
725 p->exit_signal = SIGCHLD;
727 /* If we'd notified the old parent about this child's death,
728 * also notify the new parent.
730 if (!traced && p->exit_state == EXIT_ZOMBIE &&
731 p->exit_signal != -1 && thread_group_empty(p))
732 do_notify_parent(p, p->exit_signal);
734 kill_orphaned_pgrp(p, father);
738 * When we die, we re-parent all our children.
739 * Try to give them to another thread in our thread
740 * group, and if no such member exists, give it to
741 * the child reaper process (ie "init") in our pid
744 static void forget_original_parent(struct task_struct *father)
746 struct task_struct *p, *n, *reaper = father;
747 struct list_head ptrace_dead;
749 INIT_LIST_HEAD(&ptrace_dead);
751 write_lock_irq(&tasklist_lock);
754 reaper = next_thread(reaper);
755 if (reaper == father) {
756 reaper = task_child_reaper(father);
759 } while (reaper->flags & PF_EXITING);
762 * There are only two places where our children can be:
764 * - in our child list
765 * - in our ptraced child list
767 * Search them and reparent children.
769 list_for_each_entry_safe(p, n, &father->children, sibling) {
774 /* if father isn't the real parent, then ptrace must be enabled */
775 BUG_ON(father != p->real_parent && !ptrace);
777 if (father == p->real_parent) {
778 /* reparent with a reaper, real father it's us */
779 p->real_parent = reaper;
780 reparent_thread(p, father, 0);
782 /* reparent ptraced task to its real parent */
784 if (p->exit_state == EXIT_ZOMBIE && p->exit_signal != -1 &&
785 thread_group_empty(p))
786 do_notify_parent(p, p->exit_signal);
790 * if the ptraced child is a zombie with exit_signal == -1
791 * we must collect it before we exit, or it will remain
792 * zombie forever since we prevented it from self-reap itself
793 * while it was being traced by us, to be able to see it in wait4.
795 if (unlikely(ptrace && p->exit_state == EXIT_ZOMBIE && p->exit_signal == -1))
796 list_add(&p->ptrace_list, &ptrace_dead);
799 list_for_each_entry_safe(p, n, &father->ptrace_children, ptrace_list) {
800 p->real_parent = reaper;
801 reparent_thread(p, father, 1);
804 write_unlock_irq(&tasklist_lock);
805 BUG_ON(!list_empty(&father->children));
806 BUG_ON(!list_empty(&father->ptrace_children));
808 list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_list) {
809 list_del_init(&p->ptrace_list);
816 * Send signals to all our closest relatives so that they know
817 * to properly mourn us..
819 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 && tsk->exit_signal != -1 &&
853 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
854 tsk->self_exec_id != tsk->parent_exec_id)
855 && !capable(CAP_KILL))
856 tsk->exit_signal = SIGCHLD;
859 /* If something other than our normal parent is ptracing us, then
860 * send it a SIGCHLD instead of honoring exit_signal. exit_signal
861 * only has special meaning to our real parent.
863 if (tsk->exit_signal != -1 && thread_group_empty(tsk)) {
864 int signal = tsk->parent == tsk->real_parent ? tsk->exit_signal : SIGCHLD;
865 do_notify_parent(tsk, signal);
866 } else if (tsk->ptrace) {
867 do_notify_parent(tsk, SIGCHLD);
871 if (tsk->exit_signal == -1 && likely(!tsk->ptrace))
873 tsk->exit_state = state;
875 if (thread_group_leader(tsk) &&
876 tsk->signal->notify_count < 0 &&
877 tsk->signal->group_exit_task)
878 wake_up_process(tsk->signal->group_exit_task);
880 write_unlock_irq(&tasklist_lock);
882 /* If the process is dead, release it - nobody will wait for it */
883 if (state == EXIT_DEAD)
887 #ifdef CONFIG_DEBUG_STACK_USAGE
888 static void check_stack_usage(void)
890 static DEFINE_SPINLOCK(low_water_lock);
891 static int lowest_to_date = THREAD_SIZE;
892 unsigned long *n = end_of_stack(current);
897 free = (unsigned long)n - (unsigned long)end_of_stack(current);
899 if (free >= lowest_to_date)
902 spin_lock(&low_water_lock);
903 if (free < lowest_to_date) {
904 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
906 current->comm, free);
907 lowest_to_date = free;
909 spin_unlock(&low_water_lock);
912 static inline void check_stack_usage(void) {}
915 static inline void exit_child_reaper(struct task_struct *tsk)
917 if (likely(tsk->group_leader != task_child_reaper(tsk)))
920 if (tsk->nsproxy->pid_ns == &init_pid_ns)
921 panic("Attempted to kill init!");
924 * @tsk is the last thread in the 'cgroup-init' and is exiting.
925 * Terminate all remaining processes in the namespace and reap them
926 * before exiting @tsk.
928 * Note that @tsk (last thread of cgroup-init) may not necessarily
929 * be the child-reaper (i.e main thread of cgroup-init) of the
930 * namespace i.e the child_reaper may have already exited.
932 * Even after a child_reaper exits, we let it inherit orphaned children,
933 * because, pid_ns->child_reaper remains valid as long as there is
934 * at least one living sub-thread in the cgroup init.
936 * This living sub-thread of the cgroup-init will be notified when
937 * a child inherited by the 'child-reaper' exits (do_notify_parent()
938 * uses __group_send_sig_info()). Further, when reaping child processes,
939 * do_wait() iterates over children of all living sub threads.
941 * i.e even though 'child_reaper' thread is listed as the parent of the
942 * orphaned children, any living sub-thread in the cgroup-init can
943 * perform the role of the child_reaper.
945 zap_pid_ns_processes(tsk->nsproxy->pid_ns);
948 NORET_TYPE void do_exit(long code)
950 struct task_struct *tsk = current;
953 profile_task_exit(tsk);
955 WARN_ON(atomic_read(&tsk->fs_excl));
957 if (unlikely(in_interrupt()))
958 panic("Aiee, killing interrupt handler!");
959 if (unlikely(!tsk->pid))
960 panic("Attempted to kill the idle task!");
962 if (unlikely(current->ptrace & PT_TRACE_EXIT)) {
963 current->ptrace_message = code;
964 ptrace_notify((PTRACE_EVENT_EXIT << 8) | SIGTRAP);
968 * We're taking recursive faults here in do_exit. Safest is to just
969 * leave this task alone and wait for reboot.
971 if (unlikely(tsk->flags & PF_EXITING)) {
973 "Fixing recursive fault but reboot is needed!\n");
975 * We can do this unlocked here. The futex code uses
976 * this flag just to verify whether the pi state
977 * cleanup has been done or not. In the worst case it
978 * loops once more. We pretend that the cleanup was
979 * done as there is no way to return. Either the
980 * OWNER_DIED bit is set by now or we push the blocked
981 * task into the wait for ever nirwana as well.
983 tsk->flags |= PF_EXITPIDONE;
986 set_current_state(TASK_UNINTERRUPTIBLE);
990 exit_signals(tsk); /* sets PF_EXITING */
992 * tsk->flags are checked in the futex code to protect against
993 * an exiting task cleaning up the robust pi futexes.
996 spin_unlock_wait(&tsk->pi_lock);
998 if (unlikely(in_atomic()))
999 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1000 current->comm, task_pid_nr(current),
1003 acct_update_integrals(tsk);
1005 update_hiwater_rss(tsk->mm);
1006 update_hiwater_vm(tsk->mm);
1008 group_dead = atomic_dec_and_test(&tsk->signal->live);
1010 exit_child_reaper(tsk);
1011 hrtimer_cancel(&tsk->signal->real_timer);
1012 exit_itimers(tsk->signal);
1014 acct_collect(code, group_dead);
1016 if (unlikely(tsk->robust_list))
1017 exit_robust_list(tsk);
1018 #ifdef CONFIG_COMPAT
1019 if (unlikely(tsk->compat_robust_list))
1020 compat_exit_robust_list(tsk);
1025 if (unlikely(tsk->audit_context))
1028 tsk->exit_code = code;
1029 taskstats_exit(tsk, group_dead);
1038 check_stack_usage();
1040 cgroup_exit(tsk, 1);
1043 if (group_dead && tsk->signal->leader)
1044 disassociate_ctty(1);
1046 module_put(task_thread_info(tsk)->exec_domain->module);
1048 module_put(tsk->binfmt->module);
1050 proc_exit_connector(tsk);
1051 exit_notify(tsk, group_dead);
1053 mpol_put(tsk->mempolicy);
1054 tsk->mempolicy = NULL;
1058 * This must happen late, after the PID is not
1061 if (unlikely(!list_empty(&tsk->pi_state_list)))
1062 exit_pi_state_list(tsk);
1063 if (unlikely(current->pi_state_cache))
1064 kfree(current->pi_state_cache);
1067 * Make sure we are holding no locks:
1069 debug_check_no_locks_held(tsk);
1071 * We can do this unlocked here. The futex code uses this flag
1072 * just to verify whether the pi state cleanup has been done
1073 * or not. In the worst case it loops once more.
1075 tsk->flags |= PF_EXITPIDONE;
1077 if (tsk->io_context)
1080 if (tsk->splice_pipe)
1081 __free_pipe_info(tsk->splice_pipe);
1084 /* causes final put_task_struct in finish_task_switch(). */
1085 tsk->state = TASK_DEAD;
1089 /* Avoid "noreturn function does return". */
1091 cpu_relax(); /* For when BUG is null */
1094 EXPORT_SYMBOL_GPL(do_exit);
1096 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1104 EXPORT_SYMBOL(complete_and_exit);
1106 asmlinkage long sys_exit(int error_code)
1108 do_exit((error_code&0xff)<<8);
1112 * Take down every thread in the group. This is called by fatal signals
1113 * as well as by sys_exit_group (below).
1116 do_group_exit(int exit_code)
1118 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1120 if (current->signal->flags & SIGNAL_GROUP_EXIT)
1121 exit_code = current->signal->group_exit_code;
1122 else if (!thread_group_empty(current)) {
1123 struct signal_struct *const sig = current->signal;
1124 struct sighand_struct *const sighand = current->sighand;
1125 spin_lock_irq(&sighand->siglock);
1126 if (signal_group_exit(sig))
1127 /* Another thread got here before we took the lock. */
1128 exit_code = sig->group_exit_code;
1130 sig->group_exit_code = exit_code;
1131 sig->flags = SIGNAL_GROUP_EXIT;
1132 zap_other_threads(current);
1134 spin_unlock_irq(&sighand->siglock);
1142 * this kills every thread in the thread group. Note that any externally
1143 * wait4()-ing process will get the correct exit code - even if this
1144 * thread is not the thread group leader.
1146 asmlinkage void sys_exit_group(int error_code)
1148 do_group_exit((error_code & 0xff) << 8);
1151 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1153 struct pid *pid = NULL;
1154 if (type == PIDTYPE_PID)
1155 pid = task->pids[type].pid;
1156 else if (type < PIDTYPE_MAX)
1157 pid = task->group_leader->pids[type].pid;
1161 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1162 struct task_struct *p)
1166 if (type < PIDTYPE_MAX) {
1167 if (task_pid_type(p, type) != pid)
1172 * Do not consider detached threads that are
1175 if (p->exit_signal == -1 && !p->ptrace)
1178 /* Wait for all children (clone and not) if __WALL is set;
1179 * otherwise, wait for clone children *only* if __WCLONE is
1180 * set; otherwise, wait for non-clone children *only*. (Note:
1181 * A "clone" child here is one that reports to its parent
1182 * using a signal other than SIGCHLD.) */
1183 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1184 && !(options & __WALL))
1187 err = security_task_wait(p);
1191 if (type != PIDTYPE_PID)
1193 /* This child was explicitly requested, abort */
1194 read_unlock(&tasklist_lock);
1198 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1199 int why, int status,
1200 struct siginfo __user *infop,
1201 struct rusage __user *rusagep)
1203 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1207 retval = put_user(SIGCHLD, &infop->si_signo);
1209 retval = put_user(0, &infop->si_errno);
1211 retval = put_user((short)why, &infop->si_code);
1213 retval = put_user(pid, &infop->si_pid);
1215 retval = put_user(uid, &infop->si_uid);
1217 retval = put_user(status, &infop->si_status);
1224 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1225 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1226 * the lock and this task is uninteresting. If we return nonzero, we have
1227 * released the lock and the system call should return.
1229 static int wait_task_zombie(struct task_struct *p, int noreap,
1230 struct siginfo __user *infop,
1231 int __user *stat_addr, struct rusage __user *ru)
1233 unsigned long state;
1234 int retval, status, traced;
1235 pid_t pid = task_pid_vnr(p);
1237 if (unlikely(noreap)) {
1239 int exit_code = p->exit_code;
1243 read_unlock(&tasklist_lock);
1244 if ((exit_code & 0x7f) == 0) {
1246 status = exit_code >> 8;
1248 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1249 status = exit_code & 0x7f;
1251 return wait_noreap_copyout(p, pid, uid, why,
1256 * Try to move the task's state to DEAD
1257 * only one thread is allowed to do this:
1259 state = xchg(&p->exit_state, EXIT_DEAD);
1260 if (state != EXIT_ZOMBIE) {
1261 BUG_ON(state != EXIT_DEAD);
1265 /* traced means p->ptrace, but not vice versa */
1266 traced = (p->real_parent != p->parent);
1268 if (likely(!traced)) {
1269 struct signal_struct *psig;
1270 struct signal_struct *sig;
1273 * The resource counters for the group leader are in its
1274 * own task_struct. Those for dead threads in the group
1275 * are in its signal_struct, as are those for the child
1276 * processes it has previously reaped. All these
1277 * accumulate in the parent's signal_struct c* fields.
1279 * We don't bother to take a lock here to protect these
1280 * p->signal fields, because they are only touched by
1281 * __exit_signal, which runs with tasklist_lock
1282 * write-locked anyway, and so is excluded here. We do
1283 * need to protect the access to p->parent->signal fields,
1284 * as other threads in the parent group can be right
1285 * here reaping other children at the same time.
1287 spin_lock_irq(&p->parent->sighand->siglock);
1288 psig = p->parent->signal;
1291 cputime_add(psig->cutime,
1292 cputime_add(p->utime,
1293 cputime_add(sig->utime,
1296 cputime_add(psig->cstime,
1297 cputime_add(p->stime,
1298 cputime_add(sig->stime,
1301 cputime_add(psig->cgtime,
1302 cputime_add(p->gtime,
1303 cputime_add(sig->gtime,
1306 p->min_flt + sig->min_flt + sig->cmin_flt;
1308 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1310 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1312 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1314 task_io_get_inblock(p) +
1315 sig->inblock + sig->cinblock;
1317 task_io_get_oublock(p) +
1318 sig->oublock + sig->coublock;
1319 spin_unlock_irq(&p->parent->sighand->siglock);
1323 * Now we are sure this task is interesting, and no other
1324 * thread can reap it because we set its state to EXIT_DEAD.
1326 read_unlock(&tasklist_lock);
1328 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1329 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1330 ? p->signal->group_exit_code : p->exit_code;
1331 if (!retval && stat_addr)
1332 retval = put_user(status, stat_addr);
1333 if (!retval && infop)
1334 retval = put_user(SIGCHLD, &infop->si_signo);
1335 if (!retval && infop)
1336 retval = put_user(0, &infop->si_errno);
1337 if (!retval && infop) {
1340 if ((status & 0x7f) == 0) {
1344 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1347 retval = put_user((short)why, &infop->si_code);
1349 retval = put_user(status, &infop->si_status);
1351 if (!retval && infop)
1352 retval = put_user(pid, &infop->si_pid);
1353 if (!retval && infop)
1354 retval = put_user(p->uid, &infop->si_uid);
1359 write_lock_irq(&tasklist_lock);
1360 /* We dropped tasklist, ptracer could die and untrace */
1363 * If this is not a detached task, notify the parent.
1364 * If it's still not detached after that, don't release
1367 if (p->exit_signal != -1) {
1368 do_notify_parent(p, p->exit_signal);
1369 if (p->exit_signal != -1) {
1370 p->exit_state = EXIT_ZOMBIE;
1374 write_unlock_irq(&tasklist_lock);
1383 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1384 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1385 * the lock and this task is uninteresting. If we return nonzero, we have
1386 * released the lock and the system call should return.
1388 static int wait_task_stopped(struct task_struct *p,
1389 int noreap, struct siginfo __user *infop,
1390 int __user *stat_addr, struct rusage __user *ru)
1392 int retval, exit_code, why;
1393 uid_t uid = 0; /* unneeded, required by compiler */
1397 spin_lock_irq(&p->sighand->siglock);
1399 if (unlikely(!task_is_stopped_or_traced(p)))
1402 if (!(p->ptrace & PT_PTRACED) && p->signal->group_stop_count > 0)
1404 * A group stop is in progress and this is the group leader.
1405 * We won't report until all threads have stopped.
1409 exit_code = p->exit_code;
1418 spin_unlock_irq(&p->sighand->siglock);
1423 * Now we are pretty sure this task is interesting.
1424 * Make sure it doesn't get reaped out from under us while we
1425 * give up the lock and then examine it below. We don't want to
1426 * keep holding onto the tasklist_lock while we call getrusage and
1427 * possibly take page faults for user memory.
1430 pid = task_pid_vnr(p);
1431 why = (p->ptrace & PT_PTRACED) ? CLD_TRAPPED : CLD_STOPPED;
1432 read_unlock(&tasklist_lock);
1434 if (unlikely(noreap))
1435 return wait_noreap_copyout(p, pid, uid,
1439 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1440 if (!retval && stat_addr)
1441 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1442 if (!retval && infop)
1443 retval = put_user(SIGCHLD, &infop->si_signo);
1444 if (!retval && infop)
1445 retval = put_user(0, &infop->si_errno);
1446 if (!retval && infop)
1447 retval = put_user((short)why, &infop->si_code);
1448 if (!retval && infop)
1449 retval = put_user(exit_code, &infop->si_status);
1450 if (!retval && infop)
1451 retval = put_user(pid, &infop->si_pid);
1452 if (!retval && infop)
1453 retval = put_user(uid, &infop->si_uid);
1463 * Handle do_wait work for one task in a live, non-stopped state.
1464 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1465 * the lock and this task is uninteresting. If we return nonzero, we have
1466 * released the lock and the system call should return.
1468 static int wait_task_continued(struct task_struct *p, int noreap,
1469 struct siginfo __user *infop,
1470 int __user *stat_addr, struct rusage __user *ru)
1476 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1479 spin_lock_irq(&p->sighand->siglock);
1480 /* Re-check with the lock held. */
1481 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1482 spin_unlock_irq(&p->sighand->siglock);
1486 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1487 spin_unlock_irq(&p->sighand->siglock);
1489 pid = task_pid_vnr(p);
1492 read_unlock(&tasklist_lock);
1495 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1497 if (!retval && stat_addr)
1498 retval = put_user(0xffff, stat_addr);
1502 retval = wait_noreap_copyout(p, pid, uid,
1503 CLD_CONTINUED, SIGCONT,
1505 BUG_ON(retval == 0);
1511 static long do_wait(enum pid_type type, struct pid *pid, int options,
1512 struct siginfo __user *infop, int __user *stat_addr,
1513 struct rusage __user *ru)
1515 DECLARE_WAITQUEUE(wait, current);
1516 struct task_struct *tsk;
1519 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1521 /* If there is nothing that can match our critier just get out */
1523 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1527 * We will set this flag if we see any child that might later
1528 * match our criteria, even if we are not able to reap it yet.
1531 current->state = TASK_INTERRUPTIBLE;
1532 read_lock(&tasklist_lock);
1535 struct task_struct *p;
1537 list_for_each_entry(p, &tsk->children, sibling) {
1538 int ret = eligible_child(type, pid, options, p);
1542 if (unlikely(ret < 0)) {
1544 } else if (task_is_stopped_or_traced(p)) {
1546 * It's stopped now, so it might later
1547 * continue, exit, or stop again.
1550 if (!(p->ptrace & PT_PTRACED) &&
1551 !(options & WUNTRACED))
1554 retval = wait_task_stopped(p,
1555 (options & WNOWAIT), infop,
1557 } else if (p->exit_state == EXIT_ZOMBIE &&
1558 !delay_group_leader(p)) {
1560 * We don't reap group leaders with subthreads.
1562 if (!likely(options & WEXITED))
1564 retval = wait_task_zombie(p,
1565 (options & WNOWAIT), infop,
1567 } else if (p->exit_state != EXIT_DEAD) {
1569 * It's running now, so it might later
1570 * exit, stop, or stop and then continue.
1573 if (!unlikely(options & WCONTINUED))
1575 retval = wait_task_continued(p,
1576 (options & WNOWAIT), infop,
1579 if (retval != 0) /* tasklist_lock released */
1583 list_for_each_entry(p, &tsk->ptrace_children,
1585 flag = eligible_child(type, pid, options, p);
1588 if (likely(flag > 0))
1594 if (options & __WNOTHREAD)
1596 tsk = next_thread(tsk);
1597 BUG_ON(tsk->signal != current->signal);
1598 } while (tsk != current);
1599 read_unlock(&tasklist_lock);
1602 if (options & WNOHANG)
1604 retval = -ERESTARTSYS;
1605 if (signal_pending(current))
1612 current->state = TASK_RUNNING;
1613 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1619 * For a WNOHANG return, clear out all the fields
1620 * we would set so the user can easily tell the
1624 retval = put_user(0, &infop->si_signo);
1626 retval = put_user(0, &infop->si_errno);
1628 retval = put_user(0, &infop->si_code);
1630 retval = put_user(0, &infop->si_pid);
1632 retval = put_user(0, &infop->si_uid);
1634 retval = put_user(0, &infop->si_status);
1640 asmlinkage long sys_waitid(int which, pid_t upid,
1641 struct siginfo __user *infop, int options,
1642 struct rusage __user *ru)
1644 struct pid *pid = NULL;
1648 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1650 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1663 type = PIDTYPE_PGID;
1671 if (type < PIDTYPE_MAX)
1672 pid = find_get_pid(upid);
1673 ret = do_wait(type, pid, options, infop, NULL, ru);
1676 /* avoid REGPARM breakage on x86: */
1677 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1681 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1682 int options, struct rusage __user *ru)
1684 struct pid *pid = NULL;
1688 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1689 __WNOTHREAD|__WCLONE|__WALL))
1694 else if (upid < 0) {
1695 type = PIDTYPE_PGID;
1696 pid = find_get_pid(-upid);
1697 } else if (upid == 0) {
1698 type = PIDTYPE_PGID;
1699 pid = get_pid(task_pgrp(current));
1700 } else /* upid > 0 */ {
1702 pid = find_get_pid(upid);
1705 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1708 /* avoid REGPARM breakage on x86: */
1709 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1713 #ifdef __ARCH_WANT_SYS_WAITPID
1716 * sys_waitpid() remains for compatibility. waitpid() should be
1717 * implemented by calling sys_wait4() from libc.a.
1719 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1721 return sys_wait4(pid, stat_addr, options, NULL);