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->gtime = cputime_add(sig->gtime, task_gtime(tsk));
122 sig->min_flt += tsk->min_flt;
123 sig->maj_flt += tsk->maj_flt;
124 sig->nvcsw += tsk->nvcsw;
125 sig->nivcsw += tsk->nivcsw;
126 sig->inblock += task_io_get_inblock(tsk);
127 sig->oublock += task_io_get_oublock(tsk);
128 task_io_accounting_add(&sig->ioac, &tsk->ioac);
129 sig = NULL; /* Marker for below. */
132 __unhash_process(tsk);
135 * Do this under ->siglock, we can race with another thread
136 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
138 flush_sigqueue(&tsk->pending);
142 spin_unlock(&sighand->siglock);
144 __cleanup_sighand(sighand);
145 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
147 flush_sigqueue(&sig->shared_pending);
148 taskstats_tgid_free(sig);
150 * Make sure ->signal can't go away under rq->lock,
151 * see account_group_exec_runtime().
153 task_rq_unlock_wait(tsk);
154 __cleanup_signal(sig);
158 static void delayed_put_task_struct(struct rcu_head *rhp)
160 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
162 trace_sched_process_free(tsk);
163 put_task_struct(tsk);
167 void release_task(struct task_struct * p)
169 struct task_struct *leader;
172 tracehook_prepare_release_task(p);
173 /* don't need to get the RCU readlock here - the process is dead and
174 * can't be modifying its own credentials */
175 atomic_dec(&__task_cred(p)->user->processes);
178 write_lock_irq(&tasklist_lock);
179 tracehook_finish_release_task(p);
183 * If we are the last non-leader member of the thread
184 * group, and the leader is zombie, then notify the
185 * group leader's parent process. (if it wants notification.)
188 leader = p->group_leader;
189 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
190 BUG_ON(task_detached(leader));
191 do_notify_parent(leader, leader->exit_signal);
193 * If we were the last child thread and the leader has
194 * exited already, and the leader's parent ignores SIGCHLD,
195 * then we are the one who should release the leader.
197 * do_notify_parent() will have marked it self-reaping in
200 zap_leader = task_detached(leader);
203 * This maintains the invariant that release_task()
204 * only runs on a task in EXIT_DEAD, just for sanity.
207 leader->exit_state = EXIT_DEAD;
210 write_unlock_irq(&tasklist_lock);
212 call_rcu(&p->rcu, delayed_put_task_struct);
215 if (unlikely(zap_leader))
220 * This checks not only the pgrp, but falls back on the pid if no
221 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
224 * The caller must hold rcu lock or the tasklist lock.
226 struct pid *session_of_pgrp(struct pid *pgrp)
228 struct task_struct *p;
229 struct pid *sid = NULL;
231 p = pid_task(pgrp, PIDTYPE_PGID);
233 p = pid_task(pgrp, PIDTYPE_PID);
235 sid = task_session(p);
241 * Determine if a process group is "orphaned", according to the POSIX
242 * definition in 2.2.2.52. Orphaned process groups are not to be affected
243 * by terminal-generated stop signals. Newly orphaned process groups are
244 * to receive a SIGHUP and a SIGCONT.
246 * "I ask you, have you ever known what it is to be an orphan?"
248 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
250 struct task_struct *p;
252 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
253 if ((p == ignored_task) ||
254 (p->exit_state && thread_group_empty(p)) ||
255 is_global_init(p->real_parent))
258 if (task_pgrp(p->real_parent) != pgrp &&
259 task_session(p->real_parent) == task_session(p))
261 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
266 int is_current_pgrp_orphaned(void)
270 read_lock(&tasklist_lock);
271 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
272 read_unlock(&tasklist_lock);
277 static int has_stopped_jobs(struct pid *pgrp)
280 struct task_struct *p;
282 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
283 if (!task_is_stopped(p))
287 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
292 * Check to see if any process groups have become orphaned as
293 * a result of our exiting, and if they have any stopped jobs,
294 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
297 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
299 struct pid *pgrp = task_pgrp(tsk);
300 struct task_struct *ignored_task = tsk;
303 /* exit: our father is in a different pgrp than
304 * we are and we were the only connection outside.
306 parent = tsk->real_parent;
308 /* reparent: our child is in a different pgrp than
309 * we are, and it was the only connection outside.
313 if (task_pgrp(parent) != pgrp &&
314 task_session(parent) == task_session(tsk) &&
315 will_become_orphaned_pgrp(pgrp, ignored_task) &&
316 has_stopped_jobs(pgrp)) {
317 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
318 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
323 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
325 * If a kernel thread is launched as a result of a system call, or if
326 * it ever exits, it should generally reparent itself to kthreadd so it
327 * isn't in the way of other processes and is correctly cleaned up on exit.
329 * The various task state such as scheduling policy and priority may have
330 * been inherited from a user process, so we reset them to sane values here.
332 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
334 static void reparent_to_kthreadd(void)
336 write_lock_irq(&tasklist_lock);
338 ptrace_unlink(current);
339 /* Reparent to init */
340 current->real_parent = current->parent = kthreadd_task;
341 list_move_tail(¤t->sibling, ¤t->real_parent->children);
343 /* Set the exit signal to SIGCHLD so we signal init on exit */
344 current->exit_signal = SIGCHLD;
346 if (task_nice(current) < 0)
347 set_user_nice(current, 0);
351 memcpy(current->signal->rlim, init_task.signal->rlim,
352 sizeof(current->signal->rlim));
354 atomic_inc(&init_cred.usage);
355 commit_creds(&init_cred);
356 write_unlock_irq(&tasklist_lock);
359 void __set_special_pids(struct pid *pid)
361 struct task_struct *curr = current->group_leader;
362 pid_t nr = pid_nr(pid);
364 if (task_session(curr) != pid) {
365 change_pid(curr, PIDTYPE_SID, pid);
366 set_task_session(curr, nr);
368 if (task_pgrp(curr) != pid) {
369 change_pid(curr, PIDTYPE_PGID, pid);
370 set_task_pgrp(curr, nr);
374 static void set_special_pids(struct pid *pid)
376 write_lock_irq(&tasklist_lock);
377 __set_special_pids(pid);
378 write_unlock_irq(&tasklist_lock);
382 * Let kernel threads use this to say that they
383 * allow a certain signal (since daemonize() will
384 * have disabled all of them by default).
386 int allow_signal(int sig)
388 if (!valid_signal(sig) || sig < 1)
391 spin_lock_irq(¤t->sighand->siglock);
392 sigdelset(¤t->blocked, sig);
394 /* Kernel threads handle their own signals.
395 Let the signal code know it'll be handled, so
396 that they don't get converted to SIGKILL or
397 just silently dropped */
398 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
401 spin_unlock_irq(¤t->sighand->siglock);
405 EXPORT_SYMBOL(allow_signal);
407 int disallow_signal(int sig)
409 if (!valid_signal(sig) || sig < 1)
412 spin_lock_irq(¤t->sighand->siglock);
413 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
415 spin_unlock_irq(¤t->sighand->siglock);
419 EXPORT_SYMBOL(disallow_signal);
422 * Put all the gunge required to become a kernel thread without
423 * attached user resources in one place where it belongs.
426 void daemonize(const char *name, ...)
429 struct fs_struct *fs;
432 va_start(args, name);
433 vsnprintf(current->comm, sizeof(current->comm), name, args);
437 * If we were started as result of loading a module, close all of the
438 * user space pages. We don't need them, and if we didn't close them
439 * they would be locked into memory.
443 * We don't want to have TIF_FREEZE set if the system-wide hibernation
444 * or suspend transition begins right now.
446 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
448 if (current->nsproxy != &init_nsproxy) {
449 get_nsproxy(&init_nsproxy);
450 switch_task_namespaces(current, &init_nsproxy);
452 set_special_pids(&init_struct_pid);
453 proc_clear_tty(current);
455 /* Block and flush all signals */
456 sigfillset(&blocked);
457 sigprocmask(SIG_BLOCK, &blocked, NULL);
458 flush_signals(current);
460 /* Become as one with the init task */
462 exit_fs(current); /* current->fs->count--; */
465 atomic_inc(&fs->count);
468 current->files = init_task.files;
469 atomic_inc(¤t->files->count);
471 reparent_to_kthreadd();
474 EXPORT_SYMBOL(daemonize);
476 static void close_files(struct files_struct * files)
484 * It is safe to dereference the fd table without RCU or
485 * ->file_lock because this is the last reference to the
488 fdt = files_fdtable(files);
492 if (i >= fdt->max_fds)
494 set = fdt->open_fds->fds_bits[j++];
497 struct file * file = xchg(&fdt->fd[i], NULL);
499 filp_close(file, files);
509 struct files_struct *get_files_struct(struct task_struct *task)
511 struct files_struct *files;
516 atomic_inc(&files->count);
522 void put_files_struct(struct files_struct *files)
526 if (atomic_dec_and_test(&files->count)) {
529 * Free the fd and fdset arrays if we expanded them.
530 * If the fdtable was embedded, pass files for freeing
531 * at the end of the RCU grace period. Otherwise,
532 * you can free files immediately.
534 fdt = files_fdtable(files);
535 if (fdt != &files->fdtab)
536 kmem_cache_free(files_cachep, files);
541 void reset_files_struct(struct files_struct *files)
543 struct task_struct *tsk = current;
544 struct files_struct *old;
550 put_files_struct(old);
553 void exit_files(struct task_struct *tsk)
555 struct files_struct * files = tsk->files;
561 put_files_struct(files);
565 void put_fs_struct(struct fs_struct *fs)
567 /* No need to hold fs->lock if we are killing it */
568 if (atomic_dec_and_test(&fs->count)) {
571 kmem_cache_free(fs_cachep, fs);
575 void exit_fs(struct task_struct *tsk)
577 struct fs_struct * fs = tsk->fs;
587 EXPORT_SYMBOL_GPL(exit_fs);
589 #ifdef CONFIG_MM_OWNER
591 * Task p is exiting and it owned mm, lets find a new owner for it
594 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
597 * If there are other users of the mm and the owner (us) is exiting
598 * we need to find a new owner to take on the responsibility.
600 if (atomic_read(&mm->mm_users) <= 1)
607 void mm_update_next_owner(struct mm_struct *mm)
609 struct task_struct *c, *g, *p = current;
612 if (!mm_need_new_owner(mm, p))
615 read_lock(&tasklist_lock);
617 * Search in the children
619 list_for_each_entry(c, &p->children, sibling) {
621 goto assign_new_owner;
625 * Search in the siblings
627 list_for_each_entry(c, &p->parent->children, sibling) {
629 goto assign_new_owner;
633 * Search through everything else. We should not get
636 do_each_thread(g, c) {
638 goto assign_new_owner;
639 } while_each_thread(g, c);
641 read_unlock(&tasklist_lock);
643 * We found no owner yet mm_users > 1: this implies that we are
644 * most likely racing with swapoff (try_to_unuse()) or /proc or
645 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
654 * The task_lock protects c->mm from changing.
655 * We always want mm->owner->mm == mm
659 * Delay read_unlock() till we have the task_lock()
660 * to ensure that c does not slip away underneath us
662 read_unlock(&tasklist_lock);
672 #endif /* CONFIG_MM_OWNER */
675 * Turn us into a lazy TLB process if we
678 static void exit_mm(struct task_struct * tsk)
680 struct mm_struct *mm = tsk->mm;
681 struct core_state *core_state;
687 * Serialize with any possible pending coredump.
688 * We must hold mmap_sem around checking core_state
689 * and clearing tsk->mm. The core-inducing thread
690 * will increment ->nr_threads for each thread in the
691 * group with ->mm != NULL.
693 down_read(&mm->mmap_sem);
694 core_state = mm->core_state;
696 struct core_thread self;
697 up_read(&mm->mmap_sem);
700 self.next = xchg(&core_state->dumper.next, &self);
702 * Implies mb(), the result of xchg() must be visible
703 * to core_state->dumper.
705 if (atomic_dec_and_test(&core_state->nr_threads))
706 complete(&core_state->startup);
709 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
710 if (!self.task) /* see coredump_finish() */
714 __set_task_state(tsk, TASK_RUNNING);
715 down_read(&mm->mmap_sem);
717 atomic_inc(&mm->mm_count);
718 BUG_ON(mm != tsk->active_mm);
719 /* more a memory barrier than a real lock */
722 up_read(&mm->mmap_sem);
723 enter_lazy_tlb(mm, current);
724 /* We don't want this task to be frozen prematurely */
725 clear_freeze_flag(tsk);
727 mm_update_next_owner(mm);
732 * Return nonzero if @parent's children should reap themselves.
734 * Called with write_lock_irq(&tasklist_lock) held.
736 static int ignoring_children(struct task_struct *parent)
739 struct sighand_struct *psig = parent->sighand;
741 spin_lock_irqsave(&psig->siglock, flags);
742 ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
743 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
744 spin_unlock_irqrestore(&psig->siglock, flags);
749 * Detach all tasks we were using ptrace on.
750 * Any that need to be release_task'd are put on the @dead list.
752 * Called with write_lock(&tasklist_lock) held.
754 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
756 struct task_struct *p, *n;
759 list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
762 if (p->exit_state != EXIT_ZOMBIE)
766 * If it's a zombie, our attachedness prevented normal
767 * parent notification or self-reaping. Do notification
768 * now if it would have happened earlier. If it should
769 * reap itself, add it to the @dead list. We can't call
770 * release_task() here because we already hold tasklist_lock.
772 * If it's our own child, there is no notification to do.
773 * But if our normal children self-reap, then this child
774 * was prevented by ptrace and we must reap it now.
776 if (!task_detached(p) && thread_group_empty(p)) {
777 if (!same_thread_group(p->real_parent, parent))
778 do_notify_parent(p, p->exit_signal);
781 ign = ignoring_children(parent);
787 if (task_detached(p)) {
789 * Mark it as in the process of being reaped.
791 p->exit_state = EXIT_DEAD;
792 list_add(&p->ptrace_entry, dead);
798 * Finish up exit-time ptrace cleanup.
800 * Called without locks.
802 static void ptrace_exit_finish(struct task_struct *parent,
803 struct list_head *dead)
805 struct task_struct *p, *n;
807 BUG_ON(!list_empty(&parent->ptraced));
809 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
810 list_del_init(&p->ptrace_entry);
815 static void reparent_thread(struct task_struct *p, struct task_struct *father)
817 if (p->pdeath_signal)
818 /* We already hold the tasklist_lock here. */
819 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
821 list_move_tail(&p->sibling, &p->real_parent->children);
823 /* If this is a threaded reparent there is no need to
824 * notify anyone anything has happened.
826 if (same_thread_group(p->real_parent, father))
829 /* We don't want people slaying init. */
830 if (!task_detached(p))
831 p->exit_signal = SIGCHLD;
833 /* If we'd notified the old parent about this child's death,
834 * also notify the new parent.
836 if (!ptrace_reparented(p) &&
837 p->exit_state == EXIT_ZOMBIE &&
838 !task_detached(p) && thread_group_empty(p))
839 do_notify_parent(p, p->exit_signal);
841 kill_orphaned_pgrp(p, father);
845 * When we die, we re-parent all our children.
846 * Try to give them to another thread in our thread
847 * group, and if no such member exists, give it to
848 * the child reaper process (ie "init") in our pid
851 static struct task_struct *find_new_reaper(struct task_struct *father)
853 struct pid_namespace *pid_ns = task_active_pid_ns(father);
854 struct task_struct *thread;
857 while_each_thread(father, thread) {
858 if (thread->flags & PF_EXITING)
860 if (unlikely(pid_ns->child_reaper == father))
861 pid_ns->child_reaper = thread;
865 if (unlikely(pid_ns->child_reaper == father)) {
866 write_unlock_irq(&tasklist_lock);
867 if (unlikely(pid_ns == &init_pid_ns))
868 panic("Attempted to kill init!");
870 zap_pid_ns_processes(pid_ns);
871 write_lock_irq(&tasklist_lock);
873 * We can not clear ->child_reaper or leave it alone.
874 * There may by stealth EXIT_DEAD tasks on ->children,
875 * forget_original_parent() must move them somewhere.
877 pid_ns->child_reaper = init_pid_ns.child_reaper;
880 return pid_ns->child_reaper;
883 static void forget_original_parent(struct task_struct *father)
885 struct task_struct *p, *n, *reaper;
886 LIST_HEAD(ptrace_dead);
888 write_lock_irq(&tasklist_lock);
889 reaper = find_new_reaper(father);
891 * First clean up ptrace if we were using it.
893 ptrace_exit(father, &ptrace_dead);
895 list_for_each_entry_safe(p, n, &father->children, sibling) {
896 p->real_parent = reaper;
897 if (p->parent == father) {
899 p->parent = p->real_parent;
901 reparent_thread(p, father);
904 write_unlock_irq(&tasklist_lock);
905 BUG_ON(!list_empty(&father->children));
907 ptrace_exit_finish(father, &ptrace_dead);
911 * Send signals to all our closest relatives so that they know
912 * to properly mourn us..
914 static void exit_notify(struct task_struct *tsk, int group_dead)
920 * This does two things:
922 * A. Make init inherit all the child processes
923 * B. Check to see if any process groups have become orphaned
924 * as a result of our exiting, and if they have any stopped
925 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
927 forget_original_parent(tsk);
928 exit_task_namespaces(tsk);
930 write_lock_irq(&tasklist_lock);
932 kill_orphaned_pgrp(tsk->group_leader, NULL);
934 /* Let father know we died
936 * Thread signals are configurable, but you aren't going to use
937 * that to send signals to arbitary processes.
938 * That stops right now.
940 * If the parent exec id doesn't match the exec id we saved
941 * when we started then we know the parent has changed security
944 * If our self_exec id doesn't match our parent_exec_id then
945 * we have changed execution domain as these two values started
946 * the same after a fork.
948 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
949 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
950 tsk->self_exec_id != tsk->parent_exec_id) &&
952 tsk->exit_signal = SIGCHLD;
954 signal = tracehook_notify_death(tsk, &cookie, group_dead);
956 signal = do_notify_parent(tsk, signal);
958 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
960 /* mt-exec, de_thread() is waiting for us */
961 if (thread_group_leader(tsk) &&
962 tsk->signal->group_exit_task &&
963 tsk->signal->notify_count < 0)
964 wake_up_process(tsk->signal->group_exit_task);
966 write_unlock_irq(&tasklist_lock);
968 tracehook_report_death(tsk, signal, cookie, group_dead);
970 /* If the process is dead, release it - nobody will wait for it */
971 if (signal == DEATH_REAP)
975 #ifdef CONFIG_DEBUG_STACK_USAGE
976 static void check_stack_usage(void)
978 static DEFINE_SPINLOCK(low_water_lock);
979 static int lowest_to_date = THREAD_SIZE;
980 unsigned long *n = end_of_stack(current);
985 free = (unsigned long)n - (unsigned long)end_of_stack(current);
987 if (free >= lowest_to_date)
990 spin_lock(&low_water_lock);
991 if (free < lowest_to_date) {
992 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
994 current->comm, free);
995 lowest_to_date = free;
997 spin_unlock(&low_water_lock);
1000 static inline void check_stack_usage(void) {}
1003 NORET_TYPE void do_exit(long code)
1005 struct task_struct *tsk = current;
1008 profile_task_exit(tsk);
1010 WARN_ON(atomic_read(&tsk->fs_excl));
1012 if (unlikely(in_interrupt()))
1013 panic("Aiee, killing interrupt handler!");
1014 if (unlikely(!tsk->pid))
1015 panic("Attempted to kill the idle task!");
1017 tracehook_report_exit(&code);
1020 * We're taking recursive faults here in do_exit. Safest is to just
1021 * leave this task alone and wait for reboot.
1023 if (unlikely(tsk->flags & PF_EXITING)) {
1025 "Fixing recursive fault but reboot is needed!\n");
1027 * We can do this unlocked here. The futex code uses
1028 * this flag just to verify whether the pi state
1029 * cleanup has been done or not. In the worst case it
1030 * loops once more. We pretend that the cleanup was
1031 * done as there is no way to return. Either the
1032 * OWNER_DIED bit is set by now or we push the blocked
1033 * task into the wait for ever nirwana as well.
1035 tsk->flags |= PF_EXITPIDONE;
1036 set_current_state(TASK_UNINTERRUPTIBLE);
1040 exit_signals(tsk); /* sets PF_EXITING */
1042 * tsk->flags are checked in the futex code to protect against
1043 * an exiting task cleaning up the robust pi futexes.
1046 spin_unlock_wait(&tsk->pi_lock);
1048 if (unlikely(in_atomic()))
1049 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1050 current->comm, task_pid_nr(current),
1053 acct_update_integrals(tsk);
1055 group_dead = atomic_dec_and_test(&tsk->signal->live);
1057 hrtimer_cancel(&tsk->signal->real_timer);
1058 exit_itimers(tsk->signal);
1060 acct_collect(code, group_dead);
1063 if (unlikely(tsk->audit_context))
1066 tsk->exit_code = code;
1067 taskstats_exit(tsk, group_dead);
1073 trace_sched_process_exit(tsk);
1078 check_stack_usage();
1080 cgroup_exit(tsk, 1);
1082 if (group_dead && tsk->signal->leader)
1083 disassociate_ctty(1);
1085 module_put(task_thread_info(tsk)->exec_domain->module);
1087 module_put(tsk->binfmt->module);
1089 proc_exit_connector(tsk);
1090 exit_notify(tsk, group_dead);
1092 mpol_put(tsk->mempolicy);
1093 tsk->mempolicy = NULL;
1097 * This must happen late, after the PID is not
1100 if (unlikely(!list_empty(&tsk->pi_state_list)))
1101 exit_pi_state_list(tsk);
1102 if (unlikely(current->pi_state_cache))
1103 kfree(current->pi_state_cache);
1106 * Make sure we are holding no locks:
1108 debug_check_no_locks_held(tsk);
1110 * We can do this unlocked here. The futex code uses this flag
1111 * just to verify whether the pi state cleanup has been done
1112 * or not. In the worst case it loops once more.
1114 tsk->flags |= PF_EXITPIDONE;
1116 if (tsk->io_context)
1119 if (tsk->splice_pipe)
1120 __free_pipe_info(tsk->splice_pipe);
1123 /* causes final put_task_struct in finish_task_switch(). */
1124 tsk->state = TASK_DEAD;
1127 /* Avoid "noreturn function does return". */
1129 cpu_relax(); /* For when BUG is null */
1132 EXPORT_SYMBOL_GPL(do_exit);
1134 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1142 EXPORT_SYMBOL(complete_and_exit);
1144 asmlinkage long sys_exit(int error_code)
1146 do_exit((error_code&0xff)<<8);
1150 * Take down every thread in the group. This is called by fatal signals
1151 * as well as by sys_exit_group (below).
1154 do_group_exit(int exit_code)
1156 struct signal_struct *sig = current->signal;
1158 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1160 if (signal_group_exit(sig))
1161 exit_code = sig->group_exit_code;
1162 else if (!thread_group_empty(current)) {
1163 struct sighand_struct *const sighand = current->sighand;
1164 spin_lock_irq(&sighand->siglock);
1165 if (signal_group_exit(sig))
1166 /* Another thread got here before we took the lock. */
1167 exit_code = sig->group_exit_code;
1169 sig->group_exit_code = exit_code;
1170 sig->flags = SIGNAL_GROUP_EXIT;
1171 zap_other_threads(current);
1173 spin_unlock_irq(&sighand->siglock);
1181 * this kills every thread in the thread group. Note that any externally
1182 * wait4()-ing process will get the correct exit code - even if this
1183 * thread is not the thread group leader.
1185 asmlinkage void sys_exit_group(int error_code)
1187 do_group_exit((error_code & 0xff) << 8);
1190 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1192 struct pid *pid = NULL;
1193 if (type == PIDTYPE_PID)
1194 pid = task->pids[type].pid;
1195 else if (type < PIDTYPE_MAX)
1196 pid = task->group_leader->pids[type].pid;
1200 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1201 struct task_struct *p)
1205 if (type < PIDTYPE_MAX) {
1206 if (task_pid_type(p, type) != pid)
1210 /* Wait for all children (clone and not) if __WALL is set;
1211 * otherwise, wait for clone children *only* if __WCLONE is
1212 * set; otherwise, wait for non-clone children *only*. (Note:
1213 * A "clone" child here is one that reports to its parent
1214 * using a signal other than SIGCHLD.) */
1215 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1216 && !(options & __WALL))
1219 err = security_task_wait(p);
1226 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1227 int why, int status,
1228 struct siginfo __user *infop,
1229 struct rusage __user *rusagep)
1231 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1235 retval = put_user(SIGCHLD, &infop->si_signo);
1237 retval = put_user(0, &infop->si_errno);
1239 retval = put_user((short)why, &infop->si_code);
1241 retval = put_user(pid, &infop->si_pid);
1243 retval = put_user(uid, &infop->si_uid);
1245 retval = put_user(status, &infop->si_status);
1252 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1253 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1254 * the lock and this task is uninteresting. If we return nonzero, we have
1255 * released the lock and the system call should return.
1257 static int wait_task_zombie(struct task_struct *p, int options,
1258 struct siginfo __user *infop,
1259 int __user *stat_addr, struct rusage __user *ru)
1261 unsigned long state;
1262 int retval, status, traced;
1263 pid_t pid = task_pid_vnr(p);
1264 uid_t uid = __task_cred(p)->uid;
1266 if (!likely(options & WEXITED))
1269 if (unlikely(options & WNOWAIT)) {
1270 int exit_code = p->exit_code;
1274 read_unlock(&tasklist_lock);
1275 if ((exit_code & 0x7f) == 0) {
1277 status = exit_code >> 8;
1279 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1280 status = exit_code & 0x7f;
1282 return wait_noreap_copyout(p, pid, uid, why,
1287 * Try to move the task's state to DEAD
1288 * only one thread is allowed to do this:
1290 state = xchg(&p->exit_state, EXIT_DEAD);
1291 if (state != EXIT_ZOMBIE) {
1292 BUG_ON(state != EXIT_DEAD);
1296 traced = ptrace_reparented(p);
1298 if (likely(!traced)) {
1299 struct signal_struct *psig;
1300 struct signal_struct *sig;
1301 struct task_cputime cputime;
1304 * The resource counters for the group leader are in its
1305 * own task_struct. Those for dead threads in the group
1306 * are in its signal_struct, as are those for the child
1307 * processes it has previously reaped. All these
1308 * accumulate in the parent's signal_struct c* fields.
1310 * We don't bother to take a lock here to protect these
1311 * p->signal fields, because they are only touched by
1312 * __exit_signal, which runs with tasklist_lock
1313 * write-locked anyway, and so is excluded here. We do
1314 * need to protect the access to p->parent->signal fields,
1315 * as other threads in the parent group can be right
1316 * here reaping other children at the same time.
1318 * We use thread_group_cputime() to get times for the thread
1319 * group, which consolidates times for all threads in the
1320 * group including the group leader.
1322 thread_group_cputime(p, &cputime);
1323 spin_lock_irq(&p->parent->sighand->siglock);
1324 psig = p->parent->signal;
1327 cputime_add(psig->cutime,
1328 cputime_add(cputime.utime,
1331 cputime_add(psig->cstime,
1332 cputime_add(cputime.stime,
1335 cputime_add(psig->cgtime,
1336 cputime_add(p->gtime,
1337 cputime_add(sig->gtime,
1340 p->min_flt + sig->min_flt + sig->cmin_flt;
1342 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1344 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1346 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1348 task_io_get_inblock(p) +
1349 sig->inblock + sig->cinblock;
1351 task_io_get_oublock(p) +
1352 sig->oublock + sig->coublock;
1353 task_io_accounting_add(&psig->ioac, &p->ioac);
1354 task_io_accounting_add(&psig->ioac, &sig->ioac);
1355 spin_unlock_irq(&p->parent->sighand->siglock);
1359 * Now we are sure this task is interesting, and no other
1360 * thread can reap it because we set its state to EXIT_DEAD.
1362 read_unlock(&tasklist_lock);
1364 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1365 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1366 ? p->signal->group_exit_code : p->exit_code;
1367 if (!retval && stat_addr)
1368 retval = put_user(status, stat_addr);
1369 if (!retval && infop)
1370 retval = put_user(SIGCHLD, &infop->si_signo);
1371 if (!retval && infop)
1372 retval = put_user(0, &infop->si_errno);
1373 if (!retval && infop) {
1376 if ((status & 0x7f) == 0) {
1380 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1383 retval = put_user((short)why, &infop->si_code);
1385 retval = put_user(status, &infop->si_status);
1387 if (!retval && infop)
1388 retval = put_user(pid, &infop->si_pid);
1389 if (!retval && infop)
1390 retval = put_user(uid, &infop->si_uid);
1395 write_lock_irq(&tasklist_lock);
1396 /* We dropped tasklist, ptracer could die and untrace */
1399 * If this is not a detached task, notify the parent.
1400 * If it's still not detached after that, don't release
1403 if (!task_detached(p)) {
1404 do_notify_parent(p, p->exit_signal);
1405 if (!task_detached(p)) {
1406 p->exit_state = EXIT_ZOMBIE;
1410 write_unlock_irq(&tasklist_lock);
1419 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1420 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1421 * the lock and this task is uninteresting. If we return nonzero, we have
1422 * released the lock and the system call should return.
1424 static int wait_task_stopped(int ptrace, struct task_struct *p,
1425 int options, struct siginfo __user *infop,
1426 int __user *stat_addr, struct rusage __user *ru)
1428 int retval, exit_code, why;
1429 uid_t uid = 0; /* unneeded, required by compiler */
1432 if (!(options & WUNTRACED))
1436 spin_lock_irq(&p->sighand->siglock);
1438 if (unlikely(!task_is_stopped_or_traced(p)))
1441 if (!ptrace && p->signal->group_stop_count > 0)
1443 * A group stop is in progress and this is the group leader.
1444 * We won't report until all threads have stopped.
1448 exit_code = p->exit_code;
1452 if (!unlikely(options & WNOWAIT))
1455 /* don't need the RCU readlock here as we're holding a spinlock */
1456 uid = __task_cred(p)->uid;
1458 spin_unlock_irq(&p->sighand->siglock);
1463 * Now we are pretty sure this task is interesting.
1464 * Make sure it doesn't get reaped out from under us while we
1465 * give up the lock and then examine it below. We don't want to
1466 * keep holding onto the tasklist_lock while we call getrusage and
1467 * possibly take page faults for user memory.
1470 pid = task_pid_vnr(p);
1471 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1472 read_unlock(&tasklist_lock);
1474 if (unlikely(options & WNOWAIT))
1475 return wait_noreap_copyout(p, pid, uid,
1479 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1480 if (!retval && stat_addr)
1481 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1482 if (!retval && infop)
1483 retval = put_user(SIGCHLD, &infop->si_signo);
1484 if (!retval && infop)
1485 retval = put_user(0, &infop->si_errno);
1486 if (!retval && infop)
1487 retval = put_user((short)why, &infop->si_code);
1488 if (!retval && infop)
1489 retval = put_user(exit_code, &infop->si_status);
1490 if (!retval && infop)
1491 retval = put_user(pid, &infop->si_pid);
1492 if (!retval && infop)
1493 retval = put_user(uid, &infop->si_uid);
1503 * Handle do_wait work for one task in a live, non-stopped state.
1504 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1505 * the lock and this task is uninteresting. If we return nonzero, we have
1506 * released the lock and the system call should return.
1508 static int wait_task_continued(struct task_struct *p, int options,
1509 struct siginfo __user *infop,
1510 int __user *stat_addr, struct rusage __user *ru)
1516 if (!unlikely(options & WCONTINUED))
1519 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1522 spin_lock_irq(&p->sighand->siglock);
1523 /* Re-check with the lock held. */
1524 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1525 spin_unlock_irq(&p->sighand->siglock);
1528 if (!unlikely(options & WNOWAIT))
1529 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1530 uid = __task_cred(p)->uid;
1531 spin_unlock_irq(&p->sighand->siglock);
1533 pid = task_pid_vnr(p);
1535 read_unlock(&tasklist_lock);
1538 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1540 if (!retval && stat_addr)
1541 retval = put_user(0xffff, stat_addr);
1545 retval = wait_noreap_copyout(p, pid, uid,
1546 CLD_CONTINUED, SIGCONT,
1548 BUG_ON(retval == 0);
1555 * Consider @p for a wait by @parent.
1557 * -ECHILD should be in *@notask_error before the first call.
1558 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1559 * Returns zero if the search for a child should continue;
1560 * then *@notask_error is 0 if @p is an eligible child,
1561 * or another error from security_task_wait(), or still -ECHILD.
1563 static int wait_consider_task(struct task_struct *parent, int ptrace,
1564 struct task_struct *p, int *notask_error,
1565 enum pid_type type, struct pid *pid, int options,
1566 struct siginfo __user *infop,
1567 int __user *stat_addr, struct rusage __user *ru)
1569 int ret = eligible_child(type, pid, options, p);
1573 if (unlikely(ret < 0)) {
1575 * If we have not yet seen any eligible child,
1576 * then let this error code replace -ECHILD.
1577 * A permission error will give the user a clue
1578 * to look for security policy problems, rather
1579 * than for mysterious wait bugs.
1582 *notask_error = ret;
1585 if (likely(!ptrace) && unlikely(p->ptrace)) {
1587 * This child is hidden by ptrace.
1588 * We aren't allowed to see it now, but eventually we will.
1594 if (p->exit_state == EXIT_DEAD)
1598 * We don't reap group leaders with subthreads.
1600 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1601 return wait_task_zombie(p, options, infop, stat_addr, ru);
1604 * It's stopped or running now, so it might
1605 * later continue, exit, or stop again.
1609 if (task_is_stopped_or_traced(p))
1610 return wait_task_stopped(ptrace, p, options,
1611 infop, stat_addr, ru);
1613 return wait_task_continued(p, options, infop, stat_addr, ru);
1617 * Do the work of do_wait() for one thread in the group, @tsk.
1619 * -ECHILD should be in *@notask_error before the first call.
1620 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1621 * Returns zero if the search for a child should continue; then
1622 * *@notask_error is 0 if there were any eligible children,
1623 * or another error from security_task_wait(), or still -ECHILD.
1625 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1626 enum pid_type type, struct pid *pid, int options,
1627 struct siginfo __user *infop, int __user *stat_addr,
1628 struct rusage __user *ru)
1630 struct task_struct *p;
1632 list_for_each_entry(p, &tsk->children, sibling) {
1634 * Do not consider detached threads.
1636 if (!task_detached(p)) {
1637 int ret = wait_consider_task(tsk, 0, p, notask_error,
1639 infop, stat_addr, ru);
1648 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1649 enum pid_type type, struct pid *pid, int options,
1650 struct siginfo __user *infop, int __user *stat_addr,
1651 struct rusage __user *ru)
1653 struct task_struct *p;
1656 * Traditionally we see ptrace'd stopped tasks regardless of options.
1658 options |= WUNTRACED;
1660 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1661 int ret = wait_consider_task(tsk, 1, p, notask_error,
1663 infop, stat_addr, ru);
1671 static long do_wait(enum pid_type type, struct pid *pid, int options,
1672 struct siginfo __user *infop, int __user *stat_addr,
1673 struct rusage __user *ru)
1675 DECLARE_WAITQUEUE(wait, current);
1676 struct task_struct *tsk;
1679 trace_sched_process_wait(pid);
1681 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1684 * If there is nothing that can match our critiera just get out.
1685 * We will clear @retval to zero if we see any child that might later
1686 * match our criteria, even if we are not able to reap it yet.
1689 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1692 current->state = TASK_INTERRUPTIBLE;
1693 read_lock(&tasklist_lock);
1696 int tsk_result = do_wait_thread(tsk, &retval,
1698 infop, stat_addr, ru);
1700 tsk_result = ptrace_do_wait(tsk, &retval,
1702 infop, stat_addr, ru);
1705 * tasklist_lock is unlocked and we have a final result.
1707 retval = tsk_result;
1711 if (options & __WNOTHREAD)
1713 tsk = next_thread(tsk);
1714 BUG_ON(tsk->signal != current->signal);
1715 } while (tsk != current);
1716 read_unlock(&tasklist_lock);
1718 if (!retval && !(options & WNOHANG)) {
1719 retval = -ERESTARTSYS;
1720 if (!signal_pending(current)) {
1727 current->state = TASK_RUNNING;
1728 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1734 * For a WNOHANG return, clear out all the fields
1735 * we would set so the user can easily tell the
1739 retval = put_user(0, &infop->si_signo);
1741 retval = put_user(0, &infop->si_errno);
1743 retval = put_user(0, &infop->si_code);
1745 retval = put_user(0, &infop->si_pid);
1747 retval = put_user(0, &infop->si_uid);
1749 retval = put_user(0, &infop->si_status);
1755 asmlinkage long sys_waitid(int which, pid_t upid,
1756 struct siginfo __user *infop, int options,
1757 struct rusage __user *ru)
1759 struct pid *pid = NULL;
1763 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1765 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1778 type = PIDTYPE_PGID;
1786 if (type < PIDTYPE_MAX)
1787 pid = find_get_pid(upid);
1788 ret = do_wait(type, pid, options, infop, NULL, ru);
1791 /* avoid REGPARM breakage on x86: */
1792 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1796 asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
1797 int options, struct rusage __user *ru)
1799 struct pid *pid = NULL;
1803 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1804 __WNOTHREAD|__WCLONE|__WALL))
1809 else if (upid < 0) {
1810 type = PIDTYPE_PGID;
1811 pid = find_get_pid(-upid);
1812 } else if (upid == 0) {
1813 type = PIDTYPE_PGID;
1814 pid = get_pid(task_pgrp(current));
1815 } else /* upid > 0 */ {
1817 pid = find_get_pid(upid);
1820 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1823 /* avoid REGPARM breakage on x86: */
1824 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1828 #ifdef __ARCH_WANT_SYS_WAITPID
1831 * sys_waitpid() remains for compatibility. waitpid() should be
1832 * implemented by calling sys_wait4() from libc.a.
1834 asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
1836 return sys_wait4(pid, stat_addr, options, NULL);