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/iocontext.h>
17 #include <linux/key.h>
18 #include <linux/security.h>
19 #include <linux/cpu.h>
20 #include <linux/acct.h>
21 #include <linux/tsacct_kern.h>
22 #include <linux/file.h>
23 #include <linux/fdtable.h>
24 #include <linux/binfmts.h>
25 #include <linux/nsproxy.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/ptrace.h>
28 #include <linux/profile.h>
29 #include <linux/mount.h>
30 #include <linux/proc_fs.h>
31 #include <linux/kthread.h>
32 #include <linux/mempolicy.h>
33 #include <linux/taskstats_kern.h>
34 #include <linux/delayacct.h>
35 #include <linux/freezer.h>
36 #include <linux/cgroup.h>
37 #include <linux/syscalls.h>
38 #include <linux/signal.h>
39 #include <linux/posix-timers.h>
40 #include <linux/cn_proc.h>
41 #include <linux/mutex.h>
42 #include <linux/futex.h>
43 #include <linux/pipe_fs_i.h>
44 #include <linux/audit.h> /* for audit_free() */
45 #include <linux/resource.h>
46 #include <linux/blkdev.h>
47 #include <linux/task_io_accounting_ops.h>
48 #include <linux/tracehook.h>
49 #include <linux/init_task.h>
50 #include <trace/sched.h>
52 #include <asm/uaccess.h>
53 #include <asm/unistd.h>
54 #include <asm/pgtable.h>
55 #include <asm/mmu_context.h>
56 #include "cred-internals.h"
58 DEFINE_TRACE(sched_process_free);
59 DEFINE_TRACE(sched_process_exit);
60 DEFINE_TRACE(sched_process_wait);
62 static void exit_mm(struct task_struct * tsk);
64 static inline int task_detached(struct task_struct *p)
66 return p->exit_signal == -1;
69 static void __unhash_process(struct task_struct *p)
72 detach_pid(p, PIDTYPE_PID);
73 if (thread_group_leader(p)) {
74 detach_pid(p, PIDTYPE_PGID);
75 detach_pid(p, PIDTYPE_SID);
77 list_del_rcu(&p->tasks);
78 __get_cpu_var(process_counts)--;
80 list_del_rcu(&p->thread_group);
81 list_del_init(&p->sibling);
85 * This function expects the tasklist_lock write-locked.
87 static void __exit_signal(struct task_struct *tsk)
89 struct signal_struct *sig = tsk->signal;
90 struct sighand_struct *sighand;
93 BUG_ON(!atomic_read(&sig->count));
95 sighand = rcu_dereference(tsk->sighand);
96 spin_lock(&sighand->siglock);
98 posix_cpu_timers_exit(tsk);
99 if (atomic_dec_and_test(&sig->count))
100 posix_cpu_timers_exit_group(tsk);
103 * If there is any task waiting for the group exit
106 if (sig->group_exit_task && atomic_read(&sig->count) == sig->notify_count)
107 wake_up_process(sig->group_exit_task);
109 if (tsk == sig->curr_target)
110 sig->curr_target = next_thread(tsk);
112 * Accumulate here the counters for all threads but the
113 * group leader as they die, so they can be added into
114 * the process-wide totals when those are taken.
115 * The group leader stays around as a zombie as long
116 * as there are other threads. When it gets reaped,
117 * the exit.c code will add its counts into these totals.
118 * We won't ever get here for the group leader, since it
119 * will have been the last reference on the signal_struct.
121 sig->utime = cputime_add(sig->utime, task_utime(tsk));
122 sig->stime = cputime_add(sig->stime, task_stime(tsk));
123 sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
124 sig->min_flt += tsk->min_flt;
125 sig->maj_flt += tsk->maj_flt;
126 sig->nvcsw += tsk->nvcsw;
127 sig->nivcsw += tsk->nivcsw;
128 sig->inblock += task_io_get_inblock(tsk);
129 sig->oublock += task_io_get_oublock(tsk);
130 task_io_accounting_add(&sig->ioac, &tsk->ioac);
131 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
132 sig = NULL; /* Marker for below. */
135 __unhash_process(tsk);
138 * Do this under ->siglock, we can race with another thread
139 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
141 flush_sigqueue(&tsk->pending);
145 spin_unlock(&sighand->siglock);
147 __cleanup_sighand(sighand);
148 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
150 flush_sigqueue(&sig->shared_pending);
151 taskstats_tgid_free(sig);
153 * Make sure ->signal can't go away under rq->lock,
154 * see account_group_exec_runtime().
156 task_rq_unlock_wait(tsk);
157 __cleanup_signal(sig);
161 static void delayed_put_task_struct(struct rcu_head *rhp)
163 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
165 trace_sched_process_free(tsk);
166 put_task_struct(tsk);
170 void release_task(struct task_struct * p)
172 struct task_struct *leader;
175 tracehook_prepare_release_task(p);
176 /* don't need to get the RCU readlock here - the process is dead and
177 * can't be modifying its own credentials */
178 atomic_dec(&__task_cred(p)->user->processes);
181 write_lock_irq(&tasklist_lock);
182 tracehook_finish_release_task(p);
186 * If we are the last non-leader member of the thread
187 * group, and the leader is zombie, then notify the
188 * group leader's parent process. (if it wants notification.)
191 leader = p->group_leader;
192 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
193 BUG_ON(task_detached(leader));
194 do_notify_parent(leader, leader->exit_signal);
196 * If we were the last child thread and the leader has
197 * exited already, and the leader's parent ignores SIGCHLD,
198 * then we are the one who should release the leader.
200 * do_notify_parent() will have marked it self-reaping in
203 zap_leader = task_detached(leader);
206 * This maintains the invariant that release_task()
207 * only runs on a task in EXIT_DEAD, just for sanity.
210 leader->exit_state = EXIT_DEAD;
213 write_unlock_irq(&tasklist_lock);
215 call_rcu(&p->rcu, delayed_put_task_struct);
218 if (unlikely(zap_leader))
223 * This checks not only the pgrp, but falls back on the pid if no
224 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
227 * The caller must hold rcu lock or the tasklist lock.
229 struct pid *session_of_pgrp(struct pid *pgrp)
231 struct task_struct *p;
232 struct pid *sid = NULL;
234 p = pid_task(pgrp, PIDTYPE_PGID);
236 p = pid_task(pgrp, PIDTYPE_PID);
238 sid = task_session(p);
244 * Determine if a process group is "orphaned", according to the POSIX
245 * definition in 2.2.2.52. Orphaned process groups are not to be affected
246 * by terminal-generated stop signals. Newly orphaned process groups are
247 * to receive a SIGHUP and a SIGCONT.
249 * "I ask you, have you ever known what it is to be an orphan?"
251 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
253 struct task_struct *p;
255 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
256 if ((p == ignored_task) ||
257 (p->exit_state && thread_group_empty(p)) ||
258 is_global_init(p->real_parent))
261 if (task_pgrp(p->real_parent) != pgrp &&
262 task_session(p->real_parent) == task_session(p))
264 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
269 int is_current_pgrp_orphaned(void)
273 read_lock(&tasklist_lock);
274 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
275 read_unlock(&tasklist_lock);
280 static int has_stopped_jobs(struct pid *pgrp)
283 struct task_struct *p;
285 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
286 if (!task_is_stopped(p))
290 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
295 * Check to see if any process groups have become orphaned as
296 * a result of our exiting, and if they have any stopped jobs,
297 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
300 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
302 struct pid *pgrp = task_pgrp(tsk);
303 struct task_struct *ignored_task = tsk;
306 /* exit: our father is in a different pgrp than
307 * we are and we were the only connection outside.
309 parent = tsk->real_parent;
311 /* reparent: our child is in a different pgrp than
312 * we are, and it was the only connection outside.
316 if (task_pgrp(parent) != pgrp &&
317 task_session(parent) == task_session(tsk) &&
318 will_become_orphaned_pgrp(pgrp, ignored_task) &&
319 has_stopped_jobs(pgrp)) {
320 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
321 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
326 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
328 * If a kernel thread is launched as a result of a system call, or if
329 * it ever exits, it should generally reparent itself to kthreadd so it
330 * isn't in the way of other processes and is correctly cleaned up on exit.
332 * The various task state such as scheduling policy and priority may have
333 * been inherited from a user process, so we reset them to sane values here.
335 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
337 static void reparent_to_kthreadd(void)
339 write_lock_irq(&tasklist_lock);
341 ptrace_unlink(current);
342 /* Reparent to init */
343 current->real_parent = current->parent = kthreadd_task;
344 list_move_tail(¤t->sibling, ¤t->real_parent->children);
346 /* Set the exit signal to SIGCHLD so we signal init on exit */
347 current->exit_signal = SIGCHLD;
349 if (task_nice(current) < 0)
350 set_user_nice(current, 0);
354 memcpy(current->signal->rlim, init_task.signal->rlim,
355 sizeof(current->signal->rlim));
357 atomic_inc(&init_cred.usage);
358 commit_creds(&init_cred);
359 write_unlock_irq(&tasklist_lock);
362 void __set_special_pids(struct pid *pid)
364 struct task_struct *curr = current->group_leader;
365 pid_t nr = pid_nr(pid);
367 if (task_session(curr) != pid) {
368 change_pid(curr, PIDTYPE_SID, pid);
369 set_task_session(curr, nr);
371 if (task_pgrp(curr) != pid) {
372 change_pid(curr, PIDTYPE_PGID, pid);
373 set_task_pgrp(curr, nr);
377 static void set_special_pids(struct pid *pid)
379 write_lock_irq(&tasklist_lock);
380 __set_special_pids(pid);
381 write_unlock_irq(&tasklist_lock);
385 * Let kernel threads use this to say that they
386 * allow a certain signal (since daemonize() will
387 * have disabled all of them by default).
389 int allow_signal(int sig)
391 if (!valid_signal(sig) || sig < 1)
394 spin_lock_irq(¤t->sighand->siglock);
395 sigdelset(¤t->blocked, sig);
397 /* Kernel threads handle their own signals.
398 Let the signal code know it'll be handled, so
399 that they don't get converted to SIGKILL or
400 just silently dropped */
401 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
404 spin_unlock_irq(¤t->sighand->siglock);
408 EXPORT_SYMBOL(allow_signal);
410 int disallow_signal(int sig)
412 if (!valid_signal(sig) || sig < 1)
415 spin_lock_irq(¤t->sighand->siglock);
416 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
418 spin_unlock_irq(¤t->sighand->siglock);
422 EXPORT_SYMBOL(disallow_signal);
425 * Put all the gunge required to become a kernel thread without
426 * attached user resources in one place where it belongs.
429 void daemonize(const char *name, ...)
434 va_start(args, name);
435 vsnprintf(current->comm, sizeof(current->comm), name, args);
439 * If we were started as result of loading a module, close all of the
440 * user space pages. We don't need them, and if we didn't close them
441 * they would be locked into memory.
445 * We don't want to have TIF_FREEZE set if the system-wide hibernation
446 * or suspend transition begins right now.
448 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
450 if (current->nsproxy != &init_nsproxy) {
451 get_nsproxy(&init_nsproxy);
452 switch_task_namespaces(current, &init_nsproxy);
454 set_special_pids(&init_struct_pid);
455 proc_clear_tty(current);
457 /* Block and flush all signals */
458 sigfillset(&blocked);
459 sigprocmask(SIG_BLOCK, &blocked, NULL);
460 flush_signals(current);
462 /* Become as one with the init task */
464 daemonize_fs_struct();
466 current->files = init_task.files;
467 atomic_inc(¤t->files->count);
469 reparent_to_kthreadd();
472 EXPORT_SYMBOL(daemonize);
474 static void close_files(struct files_struct * files)
482 * It is safe to dereference the fd table without RCU or
483 * ->file_lock because this is the last reference to the
486 fdt = files_fdtable(files);
490 if (i >= fdt->max_fds)
492 set = fdt->open_fds->fds_bits[j++];
495 struct file * file = xchg(&fdt->fd[i], NULL);
497 filp_close(file, files);
507 struct files_struct *get_files_struct(struct task_struct *task)
509 struct files_struct *files;
514 atomic_inc(&files->count);
520 void put_files_struct(struct files_struct *files)
524 if (atomic_dec_and_test(&files->count)) {
527 * Free the fd and fdset arrays if we expanded them.
528 * If the fdtable was embedded, pass files for freeing
529 * at the end of the RCU grace period. Otherwise,
530 * you can free files immediately.
532 fdt = files_fdtable(files);
533 if (fdt != &files->fdtab)
534 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->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 * Return nonzero if @parent's children should reap themselves.
708 * Called with write_lock_irq(&tasklist_lock) held.
710 static int ignoring_children(struct task_struct *parent)
713 struct sighand_struct *psig = parent->sighand;
715 spin_lock_irqsave(&psig->siglock, flags);
716 ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
717 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
718 spin_unlock_irqrestore(&psig->siglock, flags);
723 * Detach all tasks we were using ptrace on.
724 * Any that need to be release_task'd are put on the @dead list.
726 * Called with write_lock(&tasklist_lock) held.
728 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
730 struct task_struct *p, *n;
733 list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
736 if (p->exit_state != EXIT_ZOMBIE)
740 * If it's a zombie, our attachedness prevented normal
741 * parent notification or self-reaping. Do notification
742 * now if it would have happened earlier. If it should
743 * reap itself, add it to the @dead list. We can't call
744 * release_task() here because we already hold tasklist_lock.
746 * If it's our own child, there is no notification to do.
747 * But if our normal children self-reap, then this child
748 * was prevented by ptrace and we must reap it now.
750 if (!task_detached(p) && thread_group_empty(p)) {
751 if (!same_thread_group(p->real_parent, parent))
752 do_notify_parent(p, p->exit_signal);
755 ign = ignoring_children(parent);
761 if (task_detached(p)) {
763 * Mark it as in the process of being reaped.
765 p->exit_state = EXIT_DEAD;
766 list_add(&p->ptrace_entry, dead);
772 * Finish up exit-time ptrace cleanup.
774 * Called without locks.
776 static void ptrace_exit_finish(struct task_struct *parent,
777 struct list_head *dead)
779 struct task_struct *p, *n;
781 BUG_ON(!list_empty(&parent->ptraced));
783 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
784 list_del_init(&p->ptrace_entry);
789 static void reparent_thread(struct task_struct *p, struct task_struct *father)
791 if (p->pdeath_signal)
792 /* We already hold the tasklist_lock here. */
793 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
795 list_move_tail(&p->sibling, &p->real_parent->children);
797 /* If this is a threaded reparent there is no need to
798 * notify anyone anything has happened.
800 if (same_thread_group(p->real_parent, father))
803 /* We don't want people slaying init. */
804 if (!task_detached(p))
805 p->exit_signal = SIGCHLD;
807 /* If we'd notified the old parent about this child's death,
808 * also notify the new parent.
810 if (!ptrace_reparented(p) &&
811 p->exit_state == EXIT_ZOMBIE &&
812 !task_detached(p) && thread_group_empty(p))
813 do_notify_parent(p, p->exit_signal);
815 kill_orphaned_pgrp(p, father);
819 * When we die, we re-parent all our children.
820 * Try to give them to another thread in our thread
821 * group, and if no such member exists, give it to
822 * the child reaper process (ie "init") in our pid
825 static struct task_struct *find_new_reaper(struct task_struct *father)
827 struct pid_namespace *pid_ns = task_active_pid_ns(father);
828 struct task_struct *thread;
831 while_each_thread(father, thread) {
832 if (thread->flags & PF_EXITING)
834 if (unlikely(pid_ns->child_reaper == father))
835 pid_ns->child_reaper = thread;
839 if (unlikely(pid_ns->child_reaper == father)) {
840 write_unlock_irq(&tasklist_lock);
841 if (unlikely(pid_ns == &init_pid_ns))
842 panic("Attempted to kill init!");
844 zap_pid_ns_processes(pid_ns);
845 write_lock_irq(&tasklist_lock);
847 * We can not clear ->child_reaper or leave it alone.
848 * There may by stealth EXIT_DEAD tasks on ->children,
849 * forget_original_parent() must move them somewhere.
851 pid_ns->child_reaper = init_pid_ns.child_reaper;
854 return pid_ns->child_reaper;
857 static void forget_original_parent(struct task_struct *father)
859 struct task_struct *p, *n, *reaper;
860 LIST_HEAD(ptrace_dead);
862 write_lock_irq(&tasklist_lock);
863 reaper = find_new_reaper(father);
865 * First clean up ptrace if we were using it.
867 ptrace_exit(father, &ptrace_dead);
869 list_for_each_entry_safe(p, n, &father->children, sibling) {
870 p->real_parent = reaper;
871 if (p->parent == father) {
873 p->parent = p->real_parent;
875 reparent_thread(p, father);
878 write_unlock_irq(&tasklist_lock);
879 BUG_ON(!list_empty(&father->children));
881 ptrace_exit_finish(father, &ptrace_dead);
885 * Send signals to all our closest relatives so that they know
886 * to properly mourn us..
888 static void exit_notify(struct task_struct *tsk, int group_dead)
894 * This does two things:
896 * A. Make init inherit all the child processes
897 * B. Check to see if any process groups have become orphaned
898 * as a result of our exiting, and if they have any stopped
899 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
901 forget_original_parent(tsk);
902 exit_task_namespaces(tsk);
904 write_lock_irq(&tasklist_lock);
906 kill_orphaned_pgrp(tsk->group_leader, NULL);
908 /* Let father know we died
910 * Thread signals are configurable, but you aren't going to use
911 * that to send signals to arbitary processes.
912 * That stops right now.
914 * If the parent exec id doesn't match the exec id we saved
915 * when we started then we know the parent has changed security
918 * If our self_exec id doesn't match our parent_exec_id then
919 * we have changed execution domain as these two values started
920 * the same after a fork.
922 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
923 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
924 tsk->self_exec_id != tsk->parent_exec_id) &&
926 tsk->exit_signal = SIGCHLD;
928 signal = tracehook_notify_death(tsk, &cookie, group_dead);
930 signal = do_notify_parent(tsk, signal);
932 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
934 /* mt-exec, de_thread() is waiting for us */
935 if (thread_group_leader(tsk) &&
936 tsk->signal->group_exit_task &&
937 tsk->signal->notify_count < 0)
938 wake_up_process(tsk->signal->group_exit_task);
940 write_unlock_irq(&tasklist_lock);
942 tracehook_report_death(tsk, signal, cookie, group_dead);
944 /* If the process is dead, release it - nobody will wait for it */
945 if (signal == DEATH_REAP)
949 #ifdef CONFIG_DEBUG_STACK_USAGE
950 static void check_stack_usage(void)
952 static DEFINE_SPINLOCK(low_water_lock);
953 static int lowest_to_date = THREAD_SIZE;
956 free = stack_not_used(current);
958 if (free >= lowest_to_date)
961 spin_lock(&low_water_lock);
962 if (free < lowest_to_date) {
963 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
965 current->comm, free);
966 lowest_to_date = free;
968 spin_unlock(&low_water_lock);
971 static inline void check_stack_usage(void) {}
974 NORET_TYPE void do_exit(long code)
976 struct task_struct *tsk = current;
979 profile_task_exit(tsk);
981 WARN_ON(atomic_read(&tsk->fs_excl));
983 if (unlikely(in_interrupt()))
984 panic("Aiee, killing interrupt handler!");
985 if (unlikely(!tsk->pid))
986 panic("Attempted to kill the idle task!");
988 tracehook_report_exit(&code);
991 * We're taking recursive faults here in do_exit. Safest is to just
992 * leave this task alone and wait for reboot.
994 if (unlikely(tsk->flags & PF_EXITING)) {
996 "Fixing recursive fault but reboot is needed!\n");
998 * We can do this unlocked here. The futex code uses
999 * this flag just to verify whether the pi state
1000 * cleanup has been done or not. In the worst case it
1001 * loops once more. We pretend that the cleanup was
1002 * done as there is no way to return. Either the
1003 * OWNER_DIED bit is set by now or we push the blocked
1004 * task into the wait for ever nirwana as well.
1006 tsk->flags |= PF_EXITPIDONE;
1007 set_current_state(TASK_UNINTERRUPTIBLE);
1011 exit_signals(tsk); /* sets PF_EXITING */
1013 * tsk->flags are checked in the futex code to protect against
1014 * an exiting task cleaning up the robust pi futexes.
1017 spin_unlock_wait(&tsk->pi_lock);
1019 if (unlikely(in_atomic()))
1020 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1021 current->comm, task_pid_nr(current),
1024 acct_update_integrals(tsk);
1026 group_dead = atomic_dec_and_test(&tsk->signal->live);
1028 hrtimer_cancel(&tsk->signal->real_timer);
1029 exit_itimers(tsk->signal);
1031 acct_collect(code, group_dead);
1034 if (unlikely(tsk->audit_context))
1037 tsk->exit_code = code;
1038 taskstats_exit(tsk, group_dead);
1044 trace_sched_process_exit(tsk);
1049 check_stack_usage();
1051 cgroup_exit(tsk, 1);
1053 if (group_dead && tsk->signal->leader)
1054 disassociate_ctty(1);
1056 module_put(task_thread_info(tsk)->exec_domain->module);
1058 module_put(tsk->binfmt->module);
1060 proc_exit_connector(tsk);
1061 exit_notify(tsk, group_dead);
1063 mpol_put(tsk->mempolicy);
1064 tsk->mempolicy = NULL;
1068 * This must happen late, after the PID is not
1071 if (unlikely(!list_empty(&tsk->pi_state_list)))
1072 exit_pi_state_list(tsk);
1073 if (unlikely(current->pi_state_cache))
1074 kfree(current->pi_state_cache);
1077 * Make sure we are holding no locks:
1079 debug_check_no_locks_held(tsk);
1081 * We can do this unlocked here. The futex code uses this flag
1082 * just to verify whether the pi state cleanup has been done
1083 * or not. In the worst case it loops once more.
1085 tsk->flags |= PF_EXITPIDONE;
1087 if (tsk->io_context)
1090 if (tsk->splice_pipe)
1091 __free_pipe_info(tsk->splice_pipe);
1094 /* causes final put_task_struct in finish_task_switch(). */
1095 tsk->state = TASK_DEAD;
1098 /* Avoid "noreturn function does return". */
1100 cpu_relax(); /* For when BUG is null */
1103 EXPORT_SYMBOL_GPL(do_exit);
1105 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1113 EXPORT_SYMBOL(complete_and_exit);
1115 SYSCALL_DEFINE1(exit, int, error_code)
1117 do_exit((error_code&0xff)<<8);
1121 * Take down every thread in the group. This is called by fatal signals
1122 * as well as by sys_exit_group (below).
1125 do_group_exit(int exit_code)
1127 struct signal_struct *sig = current->signal;
1129 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1131 if (signal_group_exit(sig))
1132 exit_code = sig->group_exit_code;
1133 else if (!thread_group_empty(current)) {
1134 struct sighand_struct *const sighand = current->sighand;
1135 spin_lock_irq(&sighand->siglock);
1136 if (signal_group_exit(sig))
1137 /* Another thread got here before we took the lock. */
1138 exit_code = sig->group_exit_code;
1140 sig->group_exit_code = exit_code;
1141 sig->flags = SIGNAL_GROUP_EXIT;
1142 zap_other_threads(current);
1144 spin_unlock_irq(&sighand->siglock);
1152 * this kills every thread in the thread group. Note that any externally
1153 * wait4()-ing process will get the correct exit code - even if this
1154 * thread is not the thread group leader.
1156 SYSCALL_DEFINE1(exit_group, int, error_code)
1158 do_group_exit((error_code & 0xff) << 8);
1163 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1165 struct pid *pid = NULL;
1166 if (type == PIDTYPE_PID)
1167 pid = task->pids[type].pid;
1168 else if (type < PIDTYPE_MAX)
1169 pid = task->group_leader->pids[type].pid;
1173 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1174 struct task_struct *p)
1178 if (type < PIDTYPE_MAX) {
1179 if (task_pid_type(p, type) != pid)
1183 /* Wait for all children (clone and not) if __WALL is set;
1184 * otherwise, wait for clone children *only* if __WCLONE is
1185 * set; otherwise, wait for non-clone children *only*. (Note:
1186 * A "clone" child here is one that reports to its parent
1187 * using a signal other than SIGCHLD.) */
1188 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1189 && !(options & __WALL))
1192 err = security_task_wait(p);
1199 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1200 int why, int status,
1201 struct siginfo __user *infop,
1202 struct rusage __user *rusagep)
1204 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1208 retval = put_user(SIGCHLD, &infop->si_signo);
1210 retval = put_user(0, &infop->si_errno);
1212 retval = put_user((short)why, &infop->si_code);
1214 retval = put_user(pid, &infop->si_pid);
1216 retval = put_user(uid, &infop->si_uid);
1218 retval = put_user(status, &infop->si_status);
1225 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1226 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1227 * the lock and this task is uninteresting. If we return nonzero, we have
1228 * released the lock and the system call should return.
1230 static int wait_task_zombie(struct task_struct *p, int options,
1231 struct siginfo __user *infop,
1232 int __user *stat_addr, struct rusage __user *ru)
1234 unsigned long state;
1235 int retval, status, traced;
1236 pid_t pid = task_pid_vnr(p);
1237 uid_t uid = __task_cred(p)->uid;
1239 if (!likely(options & WEXITED))
1242 if (unlikely(options & WNOWAIT)) {
1243 int exit_code = p->exit_code;
1247 read_unlock(&tasklist_lock);
1248 if ((exit_code & 0x7f) == 0) {
1250 status = exit_code >> 8;
1252 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1253 status = exit_code & 0x7f;
1255 return wait_noreap_copyout(p, pid, uid, why,
1260 * Try to move the task's state to DEAD
1261 * only one thread is allowed to do this:
1263 state = xchg(&p->exit_state, EXIT_DEAD);
1264 if (state != EXIT_ZOMBIE) {
1265 BUG_ON(state != EXIT_DEAD);
1269 traced = ptrace_reparented(p);
1271 if (likely(!traced)) {
1272 struct signal_struct *psig;
1273 struct signal_struct *sig;
1274 struct task_cputime cputime;
1277 * The resource counters for the group leader are in its
1278 * own task_struct. Those for dead threads in the group
1279 * are in its signal_struct, as are those for the child
1280 * processes it has previously reaped. All these
1281 * accumulate in the parent's signal_struct c* fields.
1283 * We don't bother to take a lock here to protect these
1284 * p->signal fields, because they are only touched by
1285 * __exit_signal, which runs with tasklist_lock
1286 * write-locked anyway, and so is excluded here. We do
1287 * need to protect the access to p->parent->signal fields,
1288 * as other threads in the parent group can be right
1289 * here reaping other children at the same time.
1291 * We use thread_group_cputime() to get times for the thread
1292 * group, which consolidates times for all threads in the
1293 * group including the group leader.
1295 thread_group_cputime(p, &cputime);
1296 spin_lock_irq(&p->parent->sighand->siglock);
1297 psig = p->parent->signal;
1300 cputime_add(psig->cutime,
1301 cputime_add(cputime.utime,
1304 cputime_add(psig->cstime,
1305 cputime_add(cputime.stime,
1308 cputime_add(psig->cgtime,
1309 cputime_add(p->gtime,
1310 cputime_add(sig->gtime,
1313 p->min_flt + sig->min_flt + sig->cmin_flt;
1315 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1317 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1319 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1321 task_io_get_inblock(p) +
1322 sig->inblock + sig->cinblock;
1324 task_io_get_oublock(p) +
1325 sig->oublock + sig->coublock;
1326 task_io_accounting_add(&psig->ioac, &p->ioac);
1327 task_io_accounting_add(&psig->ioac, &sig->ioac);
1328 spin_unlock_irq(&p->parent->sighand->siglock);
1332 * Now we are sure this task is interesting, and no other
1333 * thread can reap it because we set its state to EXIT_DEAD.
1335 read_unlock(&tasklist_lock);
1337 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1338 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1339 ? p->signal->group_exit_code : p->exit_code;
1340 if (!retval && stat_addr)
1341 retval = put_user(status, stat_addr);
1342 if (!retval && infop)
1343 retval = put_user(SIGCHLD, &infop->si_signo);
1344 if (!retval && infop)
1345 retval = put_user(0, &infop->si_errno);
1346 if (!retval && infop) {
1349 if ((status & 0x7f) == 0) {
1353 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1356 retval = put_user((short)why, &infop->si_code);
1358 retval = put_user(status, &infop->si_status);
1360 if (!retval && infop)
1361 retval = put_user(pid, &infop->si_pid);
1362 if (!retval && infop)
1363 retval = put_user(uid, &infop->si_uid);
1368 write_lock_irq(&tasklist_lock);
1369 /* We dropped tasklist, ptracer could die and untrace */
1372 * If this is not a detached task, notify the parent.
1373 * If it's still not detached after that, don't release
1376 if (!task_detached(p)) {
1377 do_notify_parent(p, p->exit_signal);
1378 if (!task_detached(p)) {
1379 p->exit_state = EXIT_ZOMBIE;
1383 write_unlock_irq(&tasklist_lock);
1392 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1393 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1394 * the lock and this task is uninteresting. If we return nonzero, we have
1395 * released the lock and the system call should return.
1397 static int wait_task_stopped(int ptrace, struct task_struct *p,
1398 int options, struct siginfo __user *infop,
1399 int __user *stat_addr, struct rusage __user *ru)
1401 int retval, exit_code, why;
1402 uid_t uid = 0; /* unneeded, required by compiler */
1405 if (!(options & WUNTRACED))
1409 spin_lock_irq(&p->sighand->siglock);
1411 if (unlikely(!task_is_stopped_or_traced(p)))
1414 if (!ptrace && p->signal->group_stop_count > 0)
1416 * A group stop is in progress and this is the group leader.
1417 * We won't report until all threads have stopped.
1421 exit_code = p->exit_code;
1425 if (!unlikely(options & WNOWAIT))
1428 /* don't need the RCU readlock here as we're holding a spinlock */
1429 uid = __task_cred(p)->uid;
1431 spin_unlock_irq(&p->sighand->siglock);
1436 * Now we are pretty sure this task is interesting.
1437 * Make sure it doesn't get reaped out from under us while we
1438 * give up the lock and then examine it below. We don't want to
1439 * keep holding onto the tasklist_lock while we call getrusage and
1440 * possibly take page faults for user memory.
1443 pid = task_pid_vnr(p);
1444 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1445 read_unlock(&tasklist_lock);
1447 if (unlikely(options & WNOWAIT))
1448 return wait_noreap_copyout(p, pid, uid,
1452 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1453 if (!retval && stat_addr)
1454 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1455 if (!retval && infop)
1456 retval = put_user(SIGCHLD, &infop->si_signo);
1457 if (!retval && infop)
1458 retval = put_user(0, &infop->si_errno);
1459 if (!retval && infop)
1460 retval = put_user((short)why, &infop->si_code);
1461 if (!retval && infop)
1462 retval = put_user(exit_code, &infop->si_status);
1463 if (!retval && infop)
1464 retval = put_user(pid, &infop->si_pid);
1465 if (!retval && infop)
1466 retval = put_user(uid, &infop->si_uid);
1476 * Handle do_wait work for one task in a live, non-stopped state.
1477 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1478 * the lock and this task is uninteresting. If we return nonzero, we have
1479 * released the lock and the system call should return.
1481 static int wait_task_continued(struct task_struct *p, int options,
1482 struct siginfo __user *infop,
1483 int __user *stat_addr, struct rusage __user *ru)
1489 if (!unlikely(options & WCONTINUED))
1492 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1495 spin_lock_irq(&p->sighand->siglock);
1496 /* Re-check with the lock held. */
1497 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1498 spin_unlock_irq(&p->sighand->siglock);
1501 if (!unlikely(options & WNOWAIT))
1502 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1503 uid = __task_cred(p)->uid;
1504 spin_unlock_irq(&p->sighand->siglock);
1506 pid = task_pid_vnr(p);
1508 read_unlock(&tasklist_lock);
1511 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1513 if (!retval && stat_addr)
1514 retval = put_user(0xffff, stat_addr);
1518 retval = wait_noreap_copyout(p, pid, uid,
1519 CLD_CONTINUED, SIGCONT,
1521 BUG_ON(retval == 0);
1528 * Consider @p for a wait by @parent.
1530 * -ECHILD should be in *@notask_error before the first call.
1531 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1532 * Returns zero if the search for a child should continue;
1533 * then *@notask_error is 0 if @p is an eligible child,
1534 * or another error from security_task_wait(), or still -ECHILD.
1536 static int wait_consider_task(struct task_struct *parent, int ptrace,
1537 struct task_struct *p, int *notask_error,
1538 enum pid_type type, struct pid *pid, int options,
1539 struct siginfo __user *infop,
1540 int __user *stat_addr, struct rusage __user *ru)
1542 int ret = eligible_child(type, pid, options, p);
1546 if (unlikely(ret < 0)) {
1548 * If we have not yet seen any eligible child,
1549 * then let this error code replace -ECHILD.
1550 * A permission error will give the user a clue
1551 * to look for security policy problems, rather
1552 * than for mysterious wait bugs.
1555 *notask_error = ret;
1558 if (likely(!ptrace) && unlikely(p->ptrace)) {
1560 * This child is hidden by ptrace.
1561 * We aren't allowed to see it now, but eventually we will.
1567 if (p->exit_state == EXIT_DEAD)
1571 * We don't reap group leaders with subthreads.
1573 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1574 return wait_task_zombie(p, options, infop, stat_addr, ru);
1577 * It's stopped or running now, so it might
1578 * later continue, exit, or stop again.
1582 if (task_is_stopped_or_traced(p))
1583 return wait_task_stopped(ptrace, p, options,
1584 infop, stat_addr, ru);
1586 return wait_task_continued(p, options, infop, stat_addr, ru);
1590 * Do the work of do_wait() for one thread in the group, @tsk.
1592 * -ECHILD should be in *@notask_error before the first call.
1593 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1594 * Returns zero if the search for a child should continue; then
1595 * *@notask_error is 0 if there were any eligible children,
1596 * or another error from security_task_wait(), or still -ECHILD.
1598 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1599 enum pid_type type, struct pid *pid, int options,
1600 struct siginfo __user *infop, int __user *stat_addr,
1601 struct rusage __user *ru)
1603 struct task_struct *p;
1605 list_for_each_entry(p, &tsk->children, sibling) {
1607 * Do not consider detached threads.
1609 if (!task_detached(p)) {
1610 int ret = wait_consider_task(tsk, 0, p, notask_error,
1612 infop, stat_addr, ru);
1621 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1622 enum pid_type type, struct pid *pid, int options,
1623 struct siginfo __user *infop, int __user *stat_addr,
1624 struct rusage __user *ru)
1626 struct task_struct *p;
1629 * Traditionally we see ptrace'd stopped tasks regardless of options.
1631 options |= WUNTRACED;
1633 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1634 int ret = wait_consider_task(tsk, 1, p, notask_error,
1636 infop, stat_addr, ru);
1644 static long do_wait(enum pid_type type, struct pid *pid, int options,
1645 struct siginfo __user *infop, int __user *stat_addr,
1646 struct rusage __user *ru)
1648 DECLARE_WAITQUEUE(wait, current);
1649 struct task_struct *tsk;
1652 trace_sched_process_wait(pid);
1654 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1657 * If there is nothing that can match our critiera just get out.
1658 * We will clear @retval to zero if we see any child that might later
1659 * match our criteria, even if we are not able to reap it yet.
1662 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1665 current->state = TASK_INTERRUPTIBLE;
1666 read_lock(&tasklist_lock);
1669 int tsk_result = do_wait_thread(tsk, &retval,
1671 infop, stat_addr, ru);
1673 tsk_result = ptrace_do_wait(tsk, &retval,
1675 infop, stat_addr, ru);
1678 * tasklist_lock is unlocked and we have a final result.
1680 retval = tsk_result;
1684 if (options & __WNOTHREAD)
1686 tsk = next_thread(tsk);
1687 BUG_ON(tsk->signal != current->signal);
1688 } while (tsk != current);
1689 read_unlock(&tasklist_lock);
1691 if (!retval && !(options & WNOHANG)) {
1692 retval = -ERESTARTSYS;
1693 if (!signal_pending(current)) {
1700 current->state = TASK_RUNNING;
1701 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1707 * For a WNOHANG return, clear out all the fields
1708 * we would set so the user can easily tell the
1712 retval = put_user(0, &infop->si_signo);
1714 retval = put_user(0, &infop->si_errno);
1716 retval = put_user(0, &infop->si_code);
1718 retval = put_user(0, &infop->si_pid);
1720 retval = put_user(0, &infop->si_uid);
1722 retval = put_user(0, &infop->si_status);
1728 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1729 infop, int, options, struct rusage __user *, ru)
1731 struct pid *pid = NULL;
1735 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1737 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1750 type = PIDTYPE_PGID;
1758 if (type < PIDTYPE_MAX)
1759 pid = find_get_pid(upid);
1760 ret = do_wait(type, pid, options, infop, NULL, ru);
1763 /* avoid REGPARM breakage on x86: */
1764 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1768 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1769 int, options, struct rusage __user *, ru)
1771 struct pid *pid = NULL;
1775 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1776 __WNOTHREAD|__WCLONE|__WALL))
1781 else if (upid < 0) {
1782 type = PIDTYPE_PGID;
1783 pid = find_get_pid(-upid);
1784 } else if (upid == 0) {
1785 type = PIDTYPE_PGID;
1786 pid = get_pid(task_pgrp(current));
1787 } else /* upid > 0 */ {
1789 pid = find_get_pid(upid);
1792 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1795 /* avoid REGPARM breakage on x86: */
1796 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1800 #ifdef __ARCH_WANT_SYS_WAITPID
1803 * sys_waitpid() remains for compatibility. waitpid() should be
1804 * implemented by calling sys_wait4() from libc.a.
1806 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1808 return sys_wait4(pid, stat_addr, options, NULL);