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 #ifdef CONFIG_PERF_COUNTERS
166 WARN_ON_ONCE(!list_empty(&tsk->perf_counter_ctx.counter_list));
168 trace_sched_process_free(tsk);
169 put_task_struct(tsk);
173 void release_task(struct task_struct * p)
175 struct task_struct *leader;
178 tracehook_prepare_release_task(p);
179 /* don't need to get the RCU readlock here - the process is dead and
180 * can't be modifying its own credentials */
181 atomic_dec(&__task_cred(p)->user->processes);
184 write_lock_irq(&tasklist_lock);
185 tracehook_finish_release_task(p);
189 * If we are the last non-leader member of the thread
190 * group, and the leader is zombie, then notify the
191 * group leader's parent process. (if it wants notification.)
194 leader = p->group_leader;
195 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
196 BUG_ON(task_detached(leader));
197 do_notify_parent(leader, leader->exit_signal);
199 * If we were the last child thread and the leader has
200 * exited already, and the leader's parent ignores SIGCHLD,
201 * then we are the one who should release the leader.
203 * do_notify_parent() will have marked it self-reaping in
206 zap_leader = task_detached(leader);
209 * This maintains the invariant that release_task()
210 * only runs on a task in EXIT_DEAD, just for sanity.
213 leader->exit_state = EXIT_DEAD;
216 write_unlock_irq(&tasklist_lock);
218 call_rcu(&p->rcu, delayed_put_task_struct);
221 if (unlikely(zap_leader))
226 * This checks not only the pgrp, but falls back on the pid if no
227 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
230 * The caller must hold rcu lock or the tasklist lock.
232 struct pid *session_of_pgrp(struct pid *pgrp)
234 struct task_struct *p;
235 struct pid *sid = NULL;
237 p = pid_task(pgrp, PIDTYPE_PGID);
239 p = pid_task(pgrp, PIDTYPE_PID);
241 sid = task_session(p);
247 * Determine if a process group is "orphaned", according to the POSIX
248 * definition in 2.2.2.52. Orphaned process groups are not to be affected
249 * by terminal-generated stop signals. Newly orphaned process groups are
250 * to receive a SIGHUP and a SIGCONT.
252 * "I ask you, have you ever known what it is to be an orphan?"
254 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
256 struct task_struct *p;
258 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
259 if ((p == ignored_task) ||
260 (p->exit_state && thread_group_empty(p)) ||
261 is_global_init(p->real_parent))
264 if (task_pgrp(p->real_parent) != pgrp &&
265 task_session(p->real_parent) == task_session(p))
267 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
272 int is_current_pgrp_orphaned(void)
276 read_lock(&tasklist_lock);
277 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
278 read_unlock(&tasklist_lock);
283 static int has_stopped_jobs(struct pid *pgrp)
286 struct task_struct *p;
288 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
289 if (!task_is_stopped(p))
293 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
298 * Check to see if any process groups have become orphaned as
299 * a result of our exiting, and if they have any stopped jobs,
300 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
303 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
305 struct pid *pgrp = task_pgrp(tsk);
306 struct task_struct *ignored_task = tsk;
309 /* exit: our father is in a different pgrp than
310 * we are and we were the only connection outside.
312 parent = tsk->real_parent;
314 /* reparent: our child is in a different pgrp than
315 * we are, and it was the only connection outside.
319 if (task_pgrp(parent) != pgrp &&
320 task_session(parent) == task_session(tsk) &&
321 will_become_orphaned_pgrp(pgrp, ignored_task) &&
322 has_stopped_jobs(pgrp)) {
323 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
324 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
329 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
331 * If a kernel thread is launched as a result of a system call, or if
332 * it ever exits, it should generally reparent itself to kthreadd so it
333 * isn't in the way of other processes and is correctly cleaned up on exit.
335 * The various task state such as scheduling policy and priority may have
336 * been inherited from a user process, so we reset them to sane values here.
338 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
340 static void reparent_to_kthreadd(void)
342 write_lock_irq(&tasklist_lock);
344 ptrace_unlink(current);
345 /* Reparent to init */
346 current->real_parent = current->parent = kthreadd_task;
347 list_move_tail(¤t->sibling, ¤t->real_parent->children);
349 /* Set the exit signal to SIGCHLD so we signal init on exit */
350 current->exit_signal = SIGCHLD;
352 if (task_nice(current) < 0)
353 set_user_nice(current, 0);
357 memcpy(current->signal->rlim, init_task.signal->rlim,
358 sizeof(current->signal->rlim));
360 atomic_inc(&init_cred.usage);
361 commit_creds(&init_cred);
362 write_unlock_irq(&tasklist_lock);
365 void __set_special_pids(struct pid *pid)
367 struct task_struct *curr = current->group_leader;
368 pid_t nr = pid_nr(pid);
370 if (task_session(curr) != pid) {
371 change_pid(curr, PIDTYPE_SID, pid);
372 set_task_session(curr, nr);
374 if (task_pgrp(curr) != pid) {
375 change_pid(curr, PIDTYPE_PGID, pid);
376 set_task_pgrp(curr, nr);
380 static void set_special_pids(struct pid *pid)
382 write_lock_irq(&tasklist_lock);
383 __set_special_pids(pid);
384 write_unlock_irq(&tasklist_lock);
388 * Let kernel threads use this to say that they
389 * allow a certain signal (since daemonize() will
390 * have disabled all of them by default).
392 int allow_signal(int sig)
394 if (!valid_signal(sig) || sig < 1)
397 spin_lock_irq(¤t->sighand->siglock);
398 sigdelset(¤t->blocked, sig);
400 /* Kernel threads handle their own signals.
401 Let the signal code know it'll be handled, so
402 that they don't get converted to SIGKILL or
403 just silently dropped */
404 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
407 spin_unlock_irq(¤t->sighand->siglock);
411 EXPORT_SYMBOL(allow_signal);
413 int disallow_signal(int sig)
415 if (!valid_signal(sig) || sig < 1)
418 spin_lock_irq(¤t->sighand->siglock);
419 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
421 spin_unlock_irq(¤t->sighand->siglock);
425 EXPORT_SYMBOL(disallow_signal);
428 * Put all the gunge required to become a kernel thread without
429 * attached user resources in one place where it belongs.
432 void daemonize(const char *name, ...)
435 struct fs_struct *fs;
438 va_start(args, name);
439 vsnprintf(current->comm, sizeof(current->comm), name, args);
443 * If we were started as result of loading a module, close all of the
444 * user space pages. We don't need them, and if we didn't close them
445 * they would be locked into memory.
449 * We don't want to have TIF_FREEZE set if the system-wide hibernation
450 * or suspend transition begins right now.
452 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
454 if (current->nsproxy != &init_nsproxy) {
455 get_nsproxy(&init_nsproxy);
456 switch_task_namespaces(current, &init_nsproxy);
458 set_special_pids(&init_struct_pid);
459 proc_clear_tty(current);
461 /* Block and flush all signals */
462 sigfillset(&blocked);
463 sigprocmask(SIG_BLOCK, &blocked, NULL);
464 flush_signals(current);
466 /* Become as one with the init task */
468 exit_fs(current); /* current->fs->count--; */
471 atomic_inc(&fs->count);
474 current->files = init_task.files;
475 atomic_inc(¤t->files->count);
477 reparent_to_kthreadd();
480 EXPORT_SYMBOL(daemonize);
482 static void close_files(struct files_struct * files)
490 * It is safe to dereference the fd table without RCU or
491 * ->file_lock because this is the last reference to the
494 fdt = files_fdtable(files);
498 if (i >= fdt->max_fds)
500 set = fdt->open_fds->fds_bits[j++];
503 struct file * file = xchg(&fdt->fd[i], NULL);
505 filp_close(file, files);
515 struct files_struct *get_files_struct(struct task_struct *task)
517 struct files_struct *files;
522 atomic_inc(&files->count);
528 void put_files_struct(struct files_struct *files)
532 if (atomic_dec_and_test(&files->count)) {
535 * Free the fd and fdset arrays if we expanded them.
536 * If the fdtable was embedded, pass files for freeing
537 * at the end of the RCU grace period. Otherwise,
538 * you can free files immediately.
540 fdt = files_fdtable(files);
541 if (fdt != &files->fdtab)
542 kmem_cache_free(files_cachep, files);
547 void reset_files_struct(struct files_struct *files)
549 struct task_struct *tsk = current;
550 struct files_struct *old;
556 put_files_struct(old);
559 void exit_files(struct task_struct *tsk)
561 struct files_struct * files = tsk->files;
567 put_files_struct(files);
571 void put_fs_struct(struct fs_struct *fs)
573 /* No need to hold fs->lock if we are killing it */
574 if (atomic_dec_and_test(&fs->count)) {
577 kmem_cache_free(fs_cachep, fs);
581 void exit_fs(struct task_struct *tsk)
583 struct fs_struct * fs = tsk->fs;
593 EXPORT_SYMBOL_GPL(exit_fs);
595 #ifdef CONFIG_MM_OWNER
597 * Task p is exiting and it owned mm, lets find a new owner for it
600 mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
603 * If there are other users of the mm and the owner (us) is exiting
604 * we need to find a new owner to take on the responsibility.
606 if (atomic_read(&mm->mm_users) <= 1)
613 void mm_update_next_owner(struct mm_struct *mm)
615 struct task_struct *c, *g, *p = current;
618 if (!mm_need_new_owner(mm, p))
621 read_lock(&tasklist_lock);
623 * Search in the children
625 list_for_each_entry(c, &p->children, sibling) {
627 goto assign_new_owner;
631 * Search in the siblings
633 list_for_each_entry(c, &p->parent->children, sibling) {
635 goto assign_new_owner;
639 * Search through everything else. We should not get
642 do_each_thread(g, c) {
644 goto assign_new_owner;
645 } while_each_thread(g, c);
647 read_unlock(&tasklist_lock);
649 * We found no owner yet mm_users > 1: this implies that we are
650 * most likely racing with swapoff (try_to_unuse()) or /proc or
651 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
660 * The task_lock protects c->mm from changing.
661 * We always want mm->owner->mm == mm
665 * Delay read_unlock() till we have the task_lock()
666 * to ensure that c does not slip away underneath us
668 read_unlock(&tasklist_lock);
678 #endif /* CONFIG_MM_OWNER */
681 * Turn us into a lazy TLB process if we
684 static void exit_mm(struct task_struct * tsk)
686 struct mm_struct *mm = tsk->mm;
687 struct core_state *core_state;
693 * Serialize with any possible pending coredump.
694 * We must hold mmap_sem around checking core_state
695 * and clearing tsk->mm. The core-inducing thread
696 * will increment ->nr_threads for each thread in the
697 * group with ->mm != NULL.
699 down_read(&mm->mmap_sem);
700 core_state = mm->core_state;
702 struct core_thread self;
703 up_read(&mm->mmap_sem);
706 self.next = xchg(&core_state->dumper.next, &self);
708 * Implies mb(), the result of xchg() must be visible
709 * to core_state->dumper.
711 if (atomic_dec_and_test(&core_state->nr_threads))
712 complete(&core_state->startup);
715 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
716 if (!self.task) /* see coredump_finish() */
720 __set_task_state(tsk, TASK_RUNNING);
721 down_read(&mm->mmap_sem);
723 atomic_inc(&mm->mm_count);
724 BUG_ON(mm != tsk->active_mm);
725 /* more a memory barrier than a real lock */
728 up_read(&mm->mmap_sem);
729 enter_lazy_tlb(mm, current);
730 /* We don't want this task to be frozen prematurely */
731 clear_freeze_flag(tsk);
733 mm_update_next_owner(mm);
738 * Return nonzero if @parent's children should reap themselves.
740 * Called with write_lock_irq(&tasklist_lock) held.
742 static int ignoring_children(struct task_struct *parent)
745 struct sighand_struct *psig = parent->sighand;
747 spin_lock_irqsave(&psig->siglock, flags);
748 ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
749 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
750 spin_unlock_irqrestore(&psig->siglock, flags);
755 * Detach all tasks we were using ptrace on.
756 * Any that need to be release_task'd are put on the @dead list.
758 * Called with write_lock(&tasklist_lock) held.
760 static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
762 struct task_struct *p, *n;
765 list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
768 if (p->exit_state != EXIT_ZOMBIE)
772 * If it's a zombie, our attachedness prevented normal
773 * parent notification or self-reaping. Do notification
774 * now if it would have happened earlier. If it should
775 * reap itself, add it to the @dead list. We can't call
776 * release_task() here because we already hold tasklist_lock.
778 * If it's our own child, there is no notification to do.
779 * But if our normal children self-reap, then this child
780 * was prevented by ptrace and we must reap it now.
782 if (!task_detached(p) && thread_group_empty(p)) {
783 if (!same_thread_group(p->real_parent, parent))
784 do_notify_parent(p, p->exit_signal);
787 ign = ignoring_children(parent);
793 if (task_detached(p)) {
795 * Mark it as in the process of being reaped.
797 p->exit_state = EXIT_DEAD;
798 list_add(&p->ptrace_entry, dead);
804 * Finish up exit-time ptrace cleanup.
806 * Called without locks.
808 static void ptrace_exit_finish(struct task_struct *parent,
809 struct list_head *dead)
811 struct task_struct *p, *n;
813 BUG_ON(!list_empty(&parent->ptraced));
815 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
816 list_del_init(&p->ptrace_entry);
821 static void reparent_thread(struct task_struct *p, struct task_struct *father)
823 if (p->pdeath_signal)
824 /* We already hold the tasklist_lock here. */
825 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
827 list_move_tail(&p->sibling, &p->real_parent->children);
829 /* If this is a threaded reparent there is no need to
830 * notify anyone anything has happened.
832 if (same_thread_group(p->real_parent, father))
835 /* We don't want people slaying init. */
836 if (!task_detached(p))
837 p->exit_signal = SIGCHLD;
839 /* If we'd notified the old parent about this child's death,
840 * also notify the new parent.
842 if (!ptrace_reparented(p) &&
843 p->exit_state == EXIT_ZOMBIE &&
844 !task_detached(p) && thread_group_empty(p))
845 do_notify_parent(p, p->exit_signal);
847 kill_orphaned_pgrp(p, father);
851 * When we die, we re-parent all our children.
852 * Try to give them to another thread in our thread
853 * group, and if no such member exists, give it to
854 * the child reaper process (ie "init") in our pid
857 static struct task_struct *find_new_reaper(struct task_struct *father)
859 struct pid_namespace *pid_ns = task_active_pid_ns(father);
860 struct task_struct *thread;
863 while_each_thread(father, thread) {
864 if (thread->flags & PF_EXITING)
866 if (unlikely(pid_ns->child_reaper == father))
867 pid_ns->child_reaper = thread;
871 if (unlikely(pid_ns->child_reaper == father)) {
872 write_unlock_irq(&tasklist_lock);
873 if (unlikely(pid_ns == &init_pid_ns))
874 panic("Attempted to kill init!");
876 zap_pid_ns_processes(pid_ns);
877 write_lock_irq(&tasklist_lock);
879 * We can not clear ->child_reaper or leave it alone.
880 * There may by stealth EXIT_DEAD tasks on ->children,
881 * forget_original_parent() must move them somewhere.
883 pid_ns->child_reaper = init_pid_ns.child_reaper;
886 return pid_ns->child_reaper;
889 static void forget_original_parent(struct task_struct *father)
891 struct task_struct *p, *n, *reaper;
892 LIST_HEAD(ptrace_dead);
894 write_lock_irq(&tasklist_lock);
895 reaper = find_new_reaper(father);
897 * First clean up ptrace if we were using it.
899 ptrace_exit(father, &ptrace_dead);
901 list_for_each_entry_safe(p, n, &father->children, sibling) {
902 p->real_parent = reaper;
903 if (p->parent == father) {
905 p->parent = p->real_parent;
907 reparent_thread(p, father);
910 write_unlock_irq(&tasklist_lock);
911 BUG_ON(!list_empty(&father->children));
913 ptrace_exit_finish(father, &ptrace_dead);
917 * Send signals to all our closest relatives so that they know
918 * to properly mourn us..
920 static void exit_notify(struct task_struct *tsk, int group_dead)
926 * This does two things:
928 * A. Make init inherit all the child processes
929 * B. Check to see if any process groups have become orphaned
930 * as a result of our exiting, and if they have any stopped
931 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
933 forget_original_parent(tsk);
934 exit_task_namespaces(tsk);
936 write_lock_irq(&tasklist_lock);
938 kill_orphaned_pgrp(tsk->group_leader, NULL);
940 /* Let father know we died
942 * Thread signals are configurable, but you aren't going to use
943 * that to send signals to arbitary processes.
944 * That stops right now.
946 * If the parent exec id doesn't match the exec id we saved
947 * when we started then we know the parent has changed security
950 * If our self_exec id doesn't match our parent_exec_id then
951 * we have changed execution domain as these two values started
952 * the same after a fork.
954 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
955 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
956 tsk->self_exec_id != tsk->parent_exec_id) &&
958 tsk->exit_signal = SIGCHLD;
960 signal = tracehook_notify_death(tsk, &cookie, group_dead);
962 signal = do_notify_parent(tsk, signal);
964 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
966 /* mt-exec, de_thread() is waiting for us */
967 if (thread_group_leader(tsk) &&
968 tsk->signal->group_exit_task &&
969 tsk->signal->notify_count < 0)
970 wake_up_process(tsk->signal->group_exit_task);
972 write_unlock_irq(&tasklist_lock);
974 tracehook_report_death(tsk, signal, cookie, group_dead);
976 /* If the process is dead, release it - nobody will wait for it */
977 if (signal == DEATH_REAP)
981 #ifdef CONFIG_DEBUG_STACK_USAGE
982 static void check_stack_usage(void)
984 static DEFINE_SPINLOCK(low_water_lock);
985 static int lowest_to_date = THREAD_SIZE;
988 free = stack_not_used(current);
990 if (free >= lowest_to_date)
993 spin_lock(&low_water_lock);
994 if (free < lowest_to_date) {
995 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
997 current->comm, free);
998 lowest_to_date = free;
1000 spin_unlock(&low_water_lock);
1003 static inline void check_stack_usage(void) {}
1006 NORET_TYPE void do_exit(long code)
1008 struct task_struct *tsk = current;
1011 profile_task_exit(tsk);
1013 WARN_ON(atomic_read(&tsk->fs_excl));
1015 if (unlikely(in_interrupt()))
1016 panic("Aiee, killing interrupt handler!");
1017 if (unlikely(!tsk->pid))
1018 panic("Attempted to kill the idle task!");
1020 tracehook_report_exit(&code);
1023 * We're taking recursive faults here in do_exit. Safest is to just
1024 * leave this task alone and wait for reboot.
1026 if (unlikely(tsk->flags & PF_EXITING)) {
1028 "Fixing recursive fault but reboot is needed!\n");
1030 * We can do this unlocked here. The futex code uses
1031 * this flag just to verify whether the pi state
1032 * cleanup has been done or not. In the worst case it
1033 * loops once more. We pretend that the cleanup was
1034 * done as there is no way to return. Either the
1035 * OWNER_DIED bit is set by now or we push the blocked
1036 * task into the wait for ever nirwana as well.
1038 tsk->flags |= PF_EXITPIDONE;
1039 set_current_state(TASK_UNINTERRUPTIBLE);
1043 exit_signals(tsk); /* sets PF_EXITING */
1045 * tsk->flags are checked in the futex code to protect against
1046 * an exiting task cleaning up the robust pi futexes.
1049 spin_unlock_wait(&tsk->pi_lock);
1051 if (unlikely(in_atomic()))
1052 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
1053 current->comm, task_pid_nr(current),
1056 acct_update_integrals(tsk);
1058 group_dead = atomic_dec_and_test(&tsk->signal->live);
1060 hrtimer_cancel(&tsk->signal->real_timer);
1061 exit_itimers(tsk->signal);
1063 acct_collect(code, group_dead);
1066 if (unlikely(tsk->audit_context))
1069 tsk->exit_code = code;
1070 taskstats_exit(tsk, group_dead);
1076 trace_sched_process_exit(tsk);
1081 check_stack_usage();
1083 cgroup_exit(tsk, 1);
1085 if (group_dead && tsk->signal->leader)
1086 disassociate_ctty(1);
1088 module_put(task_thread_info(tsk)->exec_domain->module);
1090 module_put(tsk->binfmt->module);
1092 proc_exit_connector(tsk);
1093 exit_notify(tsk, group_dead);
1095 mpol_put(tsk->mempolicy);
1096 tsk->mempolicy = NULL;
1099 if (unlikely(!list_empty(&tsk->pi_state_list)))
1100 exit_pi_state_list(tsk);
1101 if (unlikely(current->pi_state_cache))
1102 kfree(current->pi_state_cache);
1105 * Make sure we are holding no locks:
1107 debug_check_no_locks_held(tsk);
1109 * We can do this unlocked here. The futex code uses this flag
1110 * just to verify whether the pi state cleanup has been done
1111 * or not. In the worst case it loops once more.
1113 tsk->flags |= PF_EXITPIDONE;
1115 if (tsk->io_context)
1118 if (tsk->splice_pipe)
1119 __free_pipe_info(tsk->splice_pipe);
1122 /* causes final put_task_struct in finish_task_switch(). */
1123 tsk->state = TASK_DEAD;
1126 /* Avoid "noreturn function does return". */
1128 cpu_relax(); /* For when BUG is null */
1131 EXPORT_SYMBOL_GPL(do_exit);
1133 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1141 EXPORT_SYMBOL(complete_and_exit);
1143 SYSCALL_DEFINE1(exit, int, error_code)
1145 do_exit((error_code&0xff)<<8);
1149 * Take down every thread in the group. This is called by fatal signals
1150 * as well as by sys_exit_group (below).
1153 do_group_exit(int exit_code)
1155 struct signal_struct *sig = current->signal;
1157 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1159 if (signal_group_exit(sig))
1160 exit_code = sig->group_exit_code;
1161 else if (!thread_group_empty(current)) {
1162 struct sighand_struct *const sighand = current->sighand;
1163 spin_lock_irq(&sighand->siglock);
1164 if (signal_group_exit(sig))
1165 /* Another thread got here before we took the lock. */
1166 exit_code = sig->group_exit_code;
1168 sig->group_exit_code = exit_code;
1169 sig->flags = SIGNAL_GROUP_EXIT;
1170 zap_other_threads(current);
1172 spin_unlock_irq(&sighand->siglock);
1180 * this kills every thread in the thread group. Note that any externally
1181 * wait4()-ing process will get the correct exit code - even if this
1182 * thread is not the thread group leader.
1184 SYSCALL_DEFINE1(exit_group, int, error_code)
1186 do_group_exit((error_code & 0xff) << 8);
1191 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1193 struct pid *pid = NULL;
1194 if (type == PIDTYPE_PID)
1195 pid = task->pids[type].pid;
1196 else if (type < PIDTYPE_MAX)
1197 pid = task->group_leader->pids[type].pid;
1201 static int eligible_child(enum pid_type type, struct pid *pid, int options,
1202 struct task_struct *p)
1206 if (type < PIDTYPE_MAX) {
1207 if (task_pid_type(p, type) != pid)
1211 /* Wait for all children (clone and not) if __WALL is set;
1212 * otherwise, wait for clone children *only* if __WCLONE is
1213 * set; otherwise, wait for non-clone children *only*. (Note:
1214 * A "clone" child here is one that reports to its parent
1215 * using a signal other than SIGCHLD.) */
1216 if (((p->exit_signal != SIGCHLD) ^ ((options & __WCLONE) != 0))
1217 && !(options & __WALL))
1220 err = security_task_wait(p);
1227 static int wait_noreap_copyout(struct task_struct *p, pid_t pid, uid_t uid,
1228 int why, int status,
1229 struct siginfo __user *infop,
1230 struct rusage __user *rusagep)
1232 int retval = rusagep ? getrusage(p, RUSAGE_BOTH, rusagep) : 0;
1236 retval = put_user(SIGCHLD, &infop->si_signo);
1238 retval = put_user(0, &infop->si_errno);
1240 retval = put_user((short)why, &infop->si_code);
1242 retval = put_user(pid, &infop->si_pid);
1244 retval = put_user(uid, &infop->si_uid);
1246 retval = put_user(status, &infop->si_status);
1253 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1254 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1255 * the lock and this task is uninteresting. If we return nonzero, we have
1256 * released the lock and the system call should return.
1258 static int wait_task_zombie(struct task_struct *p, int options,
1259 struct siginfo __user *infop,
1260 int __user *stat_addr, struct rusage __user *ru)
1262 unsigned long state;
1263 int retval, status, traced;
1264 pid_t pid = task_pid_vnr(p);
1265 uid_t uid = __task_cred(p)->uid;
1267 if (!likely(options & WEXITED))
1270 if (unlikely(options & WNOWAIT)) {
1271 int exit_code = p->exit_code;
1275 read_unlock(&tasklist_lock);
1276 if ((exit_code & 0x7f) == 0) {
1278 status = exit_code >> 8;
1280 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1281 status = exit_code & 0x7f;
1283 return wait_noreap_copyout(p, pid, uid, why,
1288 * Try to move the task's state to DEAD
1289 * only one thread is allowed to do this:
1291 state = xchg(&p->exit_state, EXIT_DEAD);
1292 if (state != EXIT_ZOMBIE) {
1293 BUG_ON(state != EXIT_DEAD);
1297 traced = ptrace_reparented(p);
1299 if (likely(!traced)) {
1300 struct signal_struct *psig;
1301 struct signal_struct *sig;
1302 struct task_cputime cputime;
1305 * The resource counters for the group leader are in its
1306 * own task_struct. Those for dead threads in the group
1307 * are in its signal_struct, as are those for the child
1308 * processes it has previously reaped. All these
1309 * accumulate in the parent's signal_struct c* fields.
1311 * We don't bother to take a lock here to protect these
1312 * p->signal fields, because they are only touched by
1313 * __exit_signal, which runs with tasklist_lock
1314 * write-locked anyway, and so is excluded here. We do
1315 * need to protect the access to p->parent->signal fields,
1316 * as other threads in the parent group can be right
1317 * here reaping other children at the same time.
1319 * We use thread_group_cputime() to get times for the thread
1320 * group, which consolidates times for all threads in the
1321 * group including the group leader.
1323 thread_group_cputime(p, &cputime);
1324 spin_lock_irq(&p->parent->sighand->siglock);
1325 psig = p->parent->signal;
1328 cputime_add(psig->cutime,
1329 cputime_add(cputime.utime,
1332 cputime_add(psig->cstime,
1333 cputime_add(cputime.stime,
1336 cputime_add(psig->cgtime,
1337 cputime_add(p->gtime,
1338 cputime_add(sig->gtime,
1341 p->min_flt + sig->min_flt + sig->cmin_flt;
1343 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1345 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1347 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1349 task_io_get_inblock(p) +
1350 sig->inblock + sig->cinblock;
1352 task_io_get_oublock(p) +
1353 sig->oublock + sig->coublock;
1354 task_io_accounting_add(&psig->ioac, &p->ioac);
1355 task_io_accounting_add(&psig->ioac, &sig->ioac);
1356 spin_unlock_irq(&p->parent->sighand->siglock);
1360 * Now we are sure this task is interesting, and no other
1361 * thread can reap it because we set its state to EXIT_DEAD.
1363 read_unlock(&tasklist_lock);
1366 * Flush inherited counters to the parent - before the parent
1367 * gets woken up by child-exit notifications.
1369 perf_counter_exit_task(p);
1371 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1372 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1373 ? p->signal->group_exit_code : p->exit_code;
1374 if (!retval && stat_addr)
1375 retval = put_user(status, stat_addr);
1376 if (!retval && infop)
1377 retval = put_user(SIGCHLD, &infop->si_signo);
1378 if (!retval && infop)
1379 retval = put_user(0, &infop->si_errno);
1380 if (!retval && infop) {
1383 if ((status & 0x7f) == 0) {
1387 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1390 retval = put_user((short)why, &infop->si_code);
1392 retval = put_user(status, &infop->si_status);
1394 if (!retval && infop)
1395 retval = put_user(pid, &infop->si_pid);
1396 if (!retval && infop)
1397 retval = put_user(uid, &infop->si_uid);
1402 write_lock_irq(&tasklist_lock);
1403 /* We dropped tasklist, ptracer could die and untrace */
1406 * If this is not a detached task, notify the parent.
1407 * If it's still not detached after that, don't release
1410 if (!task_detached(p)) {
1411 do_notify_parent(p, p->exit_signal);
1412 if (!task_detached(p)) {
1413 p->exit_state = EXIT_ZOMBIE;
1417 write_unlock_irq(&tasklist_lock);
1426 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1427 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1428 * the lock and this task is uninteresting. If we return nonzero, we have
1429 * released the lock and the system call should return.
1431 static int wait_task_stopped(int ptrace, struct task_struct *p,
1432 int options, struct siginfo __user *infop,
1433 int __user *stat_addr, struct rusage __user *ru)
1435 int retval, exit_code, why;
1436 uid_t uid = 0; /* unneeded, required by compiler */
1439 if (!(options & WUNTRACED))
1443 spin_lock_irq(&p->sighand->siglock);
1445 if (unlikely(!task_is_stopped_or_traced(p)))
1448 if (!ptrace && p->signal->group_stop_count > 0)
1450 * A group stop is in progress and this is the group leader.
1451 * We won't report until all threads have stopped.
1455 exit_code = p->exit_code;
1459 if (!unlikely(options & WNOWAIT))
1462 /* don't need the RCU readlock here as we're holding a spinlock */
1463 uid = __task_cred(p)->uid;
1465 spin_unlock_irq(&p->sighand->siglock);
1470 * Now we are pretty sure this task is interesting.
1471 * Make sure it doesn't get reaped out from under us while we
1472 * give up the lock and then examine it below. We don't want to
1473 * keep holding onto the tasklist_lock while we call getrusage and
1474 * possibly take page faults for user memory.
1477 pid = task_pid_vnr(p);
1478 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1479 read_unlock(&tasklist_lock);
1481 if (unlikely(options & WNOWAIT))
1482 return wait_noreap_copyout(p, pid, uid,
1486 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1487 if (!retval && stat_addr)
1488 retval = put_user((exit_code << 8) | 0x7f, stat_addr);
1489 if (!retval && infop)
1490 retval = put_user(SIGCHLD, &infop->si_signo);
1491 if (!retval && infop)
1492 retval = put_user(0, &infop->si_errno);
1493 if (!retval && infop)
1494 retval = put_user((short)why, &infop->si_code);
1495 if (!retval && infop)
1496 retval = put_user(exit_code, &infop->si_status);
1497 if (!retval && infop)
1498 retval = put_user(pid, &infop->si_pid);
1499 if (!retval && infop)
1500 retval = put_user(uid, &infop->si_uid);
1510 * Handle do_wait work for one task in a live, non-stopped state.
1511 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1512 * the lock and this task is uninteresting. If we return nonzero, we have
1513 * released the lock and the system call should return.
1515 static int wait_task_continued(struct task_struct *p, int options,
1516 struct siginfo __user *infop,
1517 int __user *stat_addr, struct rusage __user *ru)
1523 if (!unlikely(options & WCONTINUED))
1526 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1529 spin_lock_irq(&p->sighand->siglock);
1530 /* Re-check with the lock held. */
1531 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1532 spin_unlock_irq(&p->sighand->siglock);
1535 if (!unlikely(options & WNOWAIT))
1536 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1537 uid = __task_cred(p)->uid;
1538 spin_unlock_irq(&p->sighand->siglock);
1540 pid = task_pid_vnr(p);
1542 read_unlock(&tasklist_lock);
1545 retval = ru ? getrusage(p, RUSAGE_BOTH, ru) : 0;
1547 if (!retval && stat_addr)
1548 retval = put_user(0xffff, stat_addr);
1552 retval = wait_noreap_copyout(p, pid, uid,
1553 CLD_CONTINUED, SIGCONT,
1555 BUG_ON(retval == 0);
1562 * Consider @p for a wait by @parent.
1564 * -ECHILD should be in *@notask_error before the first call.
1565 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1566 * Returns zero if the search for a child should continue;
1567 * then *@notask_error is 0 if @p is an eligible child,
1568 * or another error from security_task_wait(), or still -ECHILD.
1570 static int wait_consider_task(struct task_struct *parent, int ptrace,
1571 struct task_struct *p, int *notask_error,
1572 enum pid_type type, struct pid *pid, int options,
1573 struct siginfo __user *infop,
1574 int __user *stat_addr, struct rusage __user *ru)
1576 int ret = eligible_child(type, pid, options, p);
1580 if (unlikely(ret < 0)) {
1582 * If we have not yet seen any eligible child,
1583 * then let this error code replace -ECHILD.
1584 * A permission error will give the user a clue
1585 * to look for security policy problems, rather
1586 * than for mysterious wait bugs.
1589 *notask_error = ret;
1592 if (likely(!ptrace) && unlikely(p->ptrace)) {
1594 * This child is hidden by ptrace.
1595 * We aren't allowed to see it now, but eventually we will.
1601 if (p->exit_state == EXIT_DEAD)
1605 * We don't reap group leaders with subthreads.
1607 if (p->exit_state == EXIT_ZOMBIE && !delay_group_leader(p))
1608 return wait_task_zombie(p, options, infop, stat_addr, ru);
1611 * It's stopped or running now, so it might
1612 * later continue, exit, or stop again.
1616 if (task_is_stopped_or_traced(p))
1617 return wait_task_stopped(ptrace, p, options,
1618 infop, stat_addr, ru);
1620 return wait_task_continued(p, options, infop, stat_addr, ru);
1624 * Do the work of do_wait() for one thread in the group, @tsk.
1626 * -ECHILD should be in *@notask_error before the first call.
1627 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1628 * Returns zero if the search for a child should continue; then
1629 * *@notask_error is 0 if there were any eligible children,
1630 * or another error from security_task_wait(), or still -ECHILD.
1632 static int do_wait_thread(struct task_struct *tsk, int *notask_error,
1633 enum pid_type type, struct pid *pid, int options,
1634 struct siginfo __user *infop, int __user *stat_addr,
1635 struct rusage __user *ru)
1637 struct task_struct *p;
1639 list_for_each_entry(p, &tsk->children, sibling) {
1641 * Do not consider detached threads.
1643 if (!task_detached(p)) {
1644 int ret = wait_consider_task(tsk, 0, p, notask_error,
1646 infop, stat_addr, ru);
1655 static int ptrace_do_wait(struct task_struct *tsk, int *notask_error,
1656 enum pid_type type, struct pid *pid, int options,
1657 struct siginfo __user *infop, int __user *stat_addr,
1658 struct rusage __user *ru)
1660 struct task_struct *p;
1663 * Traditionally we see ptrace'd stopped tasks regardless of options.
1665 options |= WUNTRACED;
1667 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1668 int ret = wait_consider_task(tsk, 1, p, notask_error,
1670 infop, stat_addr, ru);
1678 static long do_wait(enum pid_type type, struct pid *pid, int options,
1679 struct siginfo __user *infop, int __user *stat_addr,
1680 struct rusage __user *ru)
1682 DECLARE_WAITQUEUE(wait, current);
1683 struct task_struct *tsk;
1686 trace_sched_process_wait(pid);
1688 add_wait_queue(¤t->signal->wait_chldexit,&wait);
1691 * If there is nothing that can match our critiera just get out.
1692 * We will clear @retval to zero if we see any child that might later
1693 * match our criteria, even if we are not able to reap it yet.
1696 if ((type < PIDTYPE_MAX) && (!pid || hlist_empty(&pid->tasks[type])))
1699 current->state = TASK_INTERRUPTIBLE;
1700 read_lock(&tasklist_lock);
1703 int tsk_result = do_wait_thread(tsk, &retval,
1705 infop, stat_addr, ru);
1707 tsk_result = ptrace_do_wait(tsk, &retval,
1709 infop, stat_addr, ru);
1712 * tasklist_lock is unlocked and we have a final result.
1714 retval = tsk_result;
1718 if (options & __WNOTHREAD)
1720 tsk = next_thread(tsk);
1721 BUG_ON(tsk->signal != current->signal);
1722 } while (tsk != current);
1723 read_unlock(&tasklist_lock);
1725 if (!retval && !(options & WNOHANG)) {
1726 retval = -ERESTARTSYS;
1727 if (!signal_pending(current)) {
1734 current->state = TASK_RUNNING;
1735 remove_wait_queue(¤t->signal->wait_chldexit,&wait);
1741 * For a WNOHANG return, clear out all the fields
1742 * we would set so the user can easily tell the
1746 retval = put_user(0, &infop->si_signo);
1748 retval = put_user(0, &infop->si_errno);
1750 retval = put_user(0, &infop->si_code);
1752 retval = put_user(0, &infop->si_pid);
1754 retval = put_user(0, &infop->si_uid);
1756 retval = put_user(0, &infop->si_status);
1762 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1763 infop, int, options, struct rusage __user *, ru)
1765 struct pid *pid = NULL;
1769 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1771 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1784 type = PIDTYPE_PGID;
1792 if (type < PIDTYPE_MAX)
1793 pid = find_get_pid(upid);
1794 ret = do_wait(type, pid, options, infop, NULL, ru);
1797 /* avoid REGPARM breakage on x86: */
1798 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1802 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1803 int, options, struct rusage __user *, ru)
1805 struct pid *pid = NULL;
1809 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1810 __WNOTHREAD|__WCLONE|__WALL))
1815 else if (upid < 0) {
1816 type = PIDTYPE_PGID;
1817 pid = find_get_pid(-upid);
1818 } else if (upid == 0) {
1819 type = PIDTYPE_PGID;
1820 pid = get_pid(task_pgrp(current));
1821 } else /* upid > 0 */ {
1823 pid = find_get_pid(upid);
1826 ret = do_wait(type, pid, options | WEXITED, NULL, stat_addr, ru);
1829 /* avoid REGPARM breakage on x86: */
1830 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1834 #ifdef __ARCH_WANT_SYS_WAITPID
1837 * sys_waitpid() remains for compatibility. waitpid() should be
1838 * implemented by calling sys_wait4() from libc.a.
1840 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1842 return sys_wait4(pid, stat_addr, options, NULL);