4 * Copyright (C) 1991, 1992 Linus Torvalds
8 #include <linux/slab.h>
9 #include <linux/interrupt.h>
10 #include <linux/module.h>
11 #include <linux/capability.h>
12 #include <linux/completion.h>
13 #include <linux/personality.h>
14 #include <linux/tty.h>
15 #include <linux/iocontext.h>
16 #include <linux/key.h>
17 #include <linux/security.h>
18 #include <linux/cpu.h>
19 #include <linux/acct.h>
20 #include <linux/tsacct_kern.h>
21 #include <linux/file.h>
22 #include <linux/fdtable.h>
23 #include <linux/binfmts.h>
24 #include <linux/nsproxy.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/ptrace.h>
27 #include <linux/profile.h>
28 #include <linux/mount.h>
29 #include <linux/proc_fs.h>
30 #include <linux/kthread.h>
31 #include <linux/mempolicy.h>
32 #include <linux/taskstats_kern.h>
33 #include <linux/delayacct.h>
34 #include <linux/freezer.h>
35 #include <linux/cgroup.h>
36 #include <linux/syscalls.h>
37 #include <linux/signal.h>
38 #include <linux/posix-timers.h>
39 #include <linux/cn_proc.h>
40 #include <linux/mutex.h>
41 #include <linux/futex.h>
42 #include <linux/pipe_fs_i.h>
43 #include <linux/audit.h> /* for audit_free() */
44 #include <linux/resource.h>
45 #include <linux/blkdev.h>
46 #include <linux/task_io_accounting_ops.h>
47 #include <linux/tracehook.h>
48 #include <linux/fs_struct.h>
49 #include <linux/init_task.h>
50 #include <linux/perf_event.h>
51 #include <trace/events/sched.h>
52 #include <linux/hw_breakpoint.h>
53 #include <linux/oom.h>
55 #include <asm/uaccess.h>
56 #include <asm/unistd.h>
57 #include <asm/pgtable.h>
58 #include <asm/mmu_context.h>
60 static void exit_mm(struct task_struct * tsk);
62 static void __unhash_process(struct task_struct *p, bool group_dead)
65 detach_pid(p, PIDTYPE_PID);
67 detach_pid(p, PIDTYPE_PGID);
68 detach_pid(p, PIDTYPE_SID);
70 list_del_rcu(&p->tasks);
71 list_del_init(&p->sibling);
72 __this_cpu_dec(process_counts);
74 list_del_rcu(&p->thread_group);
78 * This function expects the tasklist_lock write-locked.
80 static void __exit_signal(struct task_struct *tsk)
82 struct signal_struct *sig = tsk->signal;
83 bool group_dead = thread_group_leader(tsk);
84 struct sighand_struct *sighand;
85 struct tty_struct *uninitialized_var(tty);
87 sighand = rcu_dereference_check(tsk->sighand,
88 lockdep_tasklist_lock_is_held());
89 spin_lock(&sighand->siglock);
91 posix_cpu_timers_exit(tsk);
93 posix_cpu_timers_exit_group(tsk);
98 * This can only happen if the caller is de_thread().
99 * FIXME: this is the temporary hack, we should teach
100 * posix-cpu-timers to handle this case correctly.
102 if (unlikely(has_group_leader_pid(tsk)))
103 posix_cpu_timers_exit_group(tsk);
106 * If there is any task waiting for the group exit
109 if (sig->notify_count > 0 && !--sig->notify_count)
110 wake_up_process(sig->group_exit_task);
112 if (tsk == sig->curr_target)
113 sig->curr_target = next_thread(tsk);
115 * Accumulate here the counters for all threads but the
116 * group leader as they die, so they can be added into
117 * the process-wide totals when those are taken.
118 * The group leader stays around as a zombie as long
119 * as there are other threads. When it gets reaped,
120 * the exit.c code will add its counts into these totals.
121 * We won't ever get here for the group leader, since it
122 * will have been the last reference on the signal_struct.
124 sig->utime = cputime_add(sig->utime, tsk->utime);
125 sig->stime = cputime_add(sig->stime, tsk->stime);
126 sig->gtime = cputime_add(sig->gtime, tsk->gtime);
127 sig->min_flt += tsk->min_flt;
128 sig->maj_flt += tsk->maj_flt;
129 sig->nvcsw += tsk->nvcsw;
130 sig->nivcsw += tsk->nivcsw;
131 sig->inblock += task_io_get_inblock(tsk);
132 sig->oublock += task_io_get_oublock(tsk);
133 task_io_accounting_add(&sig->ioac, &tsk->ioac);
134 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
138 __unhash_process(tsk, group_dead);
141 * Do this under ->siglock, we can race with another thread
142 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
144 flush_sigqueue(&tsk->pending);
146 spin_unlock(&sighand->siglock);
148 __cleanup_sighand(sighand);
149 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
151 flush_sigqueue(&sig->shared_pending);
156 static void delayed_put_task_struct(struct rcu_head *rhp)
158 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
160 perf_event_delayed_put(tsk);
161 trace_sched_process_free(tsk);
162 put_task_struct(tsk);
166 void release_task(struct task_struct * p)
168 struct task_struct *leader;
171 tracehook_prepare_release_task(p);
172 /* don't need to get the RCU readlock here - the process is dead and
173 * can't be modifying its own credentials. But shut RCU-lockdep up */
175 atomic_dec(&__task_cred(p)->user->processes);
180 write_lock_irq(&tasklist_lock);
181 tracehook_finish_release_task(p);
185 * If we are the last non-leader member of the thread
186 * group, and the leader is zombie, then notify the
187 * group leader's parent process. (if it wants notification.)
190 leader = p->group_leader;
191 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
192 BUG_ON(task_detached(leader));
193 do_notify_parent(leader, leader->exit_signal);
195 * If we were the last child thread and the leader has
196 * exited already, and the leader's parent ignores SIGCHLD,
197 * then we are the one who should release the leader.
199 * do_notify_parent() will have marked it self-reaping in
202 zap_leader = task_detached(leader);
205 * This maintains the invariant that release_task()
206 * only runs on a task in EXIT_DEAD, just for sanity.
209 leader->exit_state = EXIT_DEAD;
212 write_unlock_irq(&tasklist_lock);
214 call_rcu(&p->rcu, delayed_put_task_struct);
217 if (unlikely(zap_leader))
222 * This checks not only the pgrp, but falls back on the pid if no
223 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
226 * The caller must hold rcu lock or the tasklist lock.
228 struct pid *session_of_pgrp(struct pid *pgrp)
230 struct task_struct *p;
231 struct pid *sid = NULL;
233 p = pid_task(pgrp, PIDTYPE_PGID);
235 p = pid_task(pgrp, PIDTYPE_PID);
237 sid = task_session(p);
243 * Determine if a process group is "orphaned", according to the POSIX
244 * definition in 2.2.2.52. Orphaned process groups are not to be affected
245 * by terminal-generated stop signals. Newly orphaned process groups are
246 * to receive a SIGHUP and a SIGCONT.
248 * "I ask you, have you ever known what it is to be an orphan?"
250 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
252 struct task_struct *p;
254 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
255 if ((p == ignored_task) ||
256 (p->exit_state && thread_group_empty(p)) ||
257 is_global_init(p->real_parent))
260 if (task_pgrp(p->real_parent) != pgrp &&
261 task_session(p->real_parent) == task_session(p))
263 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
268 int is_current_pgrp_orphaned(void)
272 read_lock(&tasklist_lock);
273 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
274 read_unlock(&tasklist_lock);
279 static int has_stopped_jobs(struct pid *pgrp)
282 struct task_struct *p;
284 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
285 if (!task_is_stopped(p))
289 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
294 * Check to see if any process groups have become orphaned as
295 * a result of our exiting, and if they have any stopped jobs,
296 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
299 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
301 struct pid *pgrp = task_pgrp(tsk);
302 struct task_struct *ignored_task = tsk;
305 /* exit: our father is in a different pgrp than
306 * we are and we were the only connection outside.
308 parent = tsk->real_parent;
310 /* reparent: our child is in a different pgrp than
311 * we are, and it was the only connection outside.
315 if (task_pgrp(parent) != pgrp &&
316 task_session(parent) == task_session(tsk) &&
317 will_become_orphaned_pgrp(pgrp, ignored_task) &&
318 has_stopped_jobs(pgrp)) {
319 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
320 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
325 * reparent_to_kthreadd - Reparent the calling kernel thread to kthreadd
327 * If a kernel thread is launched as a result of a system call, or if
328 * it ever exits, it should generally reparent itself to kthreadd so it
329 * isn't in the way of other processes and is correctly cleaned up on exit.
331 * The various task state such as scheduling policy and priority may have
332 * been inherited from a user process, so we reset them to sane values here.
334 * NOTE that reparent_to_kthreadd() gives the caller full capabilities.
336 static void reparent_to_kthreadd(void)
338 write_lock_irq(&tasklist_lock);
340 ptrace_unlink(current);
341 /* Reparent to init */
342 current->real_parent = current->parent = kthreadd_task;
343 list_move_tail(¤t->sibling, ¤t->real_parent->children);
345 /* Set the exit signal to SIGCHLD so we signal init on exit */
346 current->exit_signal = SIGCHLD;
348 if (task_nice(current) < 0)
349 set_user_nice(current, 0);
353 memcpy(current->signal->rlim, init_task.signal->rlim,
354 sizeof(current->signal->rlim));
356 atomic_inc(&init_cred.usage);
357 commit_creds(&init_cred);
358 write_unlock_irq(&tasklist_lock);
361 void __set_special_pids(struct pid *pid)
363 struct task_struct *curr = current->group_leader;
365 if (task_session(curr) != pid)
366 change_pid(curr, PIDTYPE_SID, pid);
368 if (task_pgrp(curr) != pid)
369 change_pid(curr, PIDTYPE_PGID, pid);
372 static void set_special_pids(struct pid *pid)
374 write_lock_irq(&tasklist_lock);
375 __set_special_pids(pid);
376 write_unlock_irq(&tasklist_lock);
380 * Let kernel threads use this to say that they allow a certain signal.
381 * Must not be used if kthread was cloned with CLONE_SIGHAND.
383 int allow_signal(int sig)
385 if (!valid_signal(sig) || sig < 1)
388 spin_lock_irq(¤t->sighand->siglock);
389 /* This is only needed for daemonize()'ed kthreads */
390 sigdelset(¤t->blocked, sig);
392 * Kernel threads handle their own signals. Let the signal code
393 * know it'll be handled, so that they don't get converted to
394 * SIGKILL or just silently dropped.
396 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
398 spin_unlock_irq(¤t->sighand->siglock);
402 EXPORT_SYMBOL(allow_signal);
404 int disallow_signal(int sig)
406 if (!valid_signal(sig) || sig < 1)
409 spin_lock_irq(¤t->sighand->siglock);
410 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
412 spin_unlock_irq(¤t->sighand->siglock);
416 EXPORT_SYMBOL(disallow_signal);
419 * Put all the gunge required to become a kernel thread without
420 * attached user resources in one place where it belongs.
423 void daemonize(const char *name, ...)
428 va_start(args, name);
429 vsnprintf(current->comm, sizeof(current->comm), name, args);
433 * If we were started as result of loading a module, close all of the
434 * user space pages. We don't need them, and if we didn't close them
435 * they would be locked into memory.
439 * We don't want to have TIF_FREEZE set if the system-wide hibernation
440 * or suspend transition begins right now.
442 current->flags |= (PF_NOFREEZE | PF_KTHREAD);
444 if (current->nsproxy != &init_nsproxy) {
445 get_nsproxy(&init_nsproxy);
446 switch_task_namespaces(current, &init_nsproxy);
448 set_special_pids(&init_struct_pid);
449 proc_clear_tty(current);
451 /* Block and flush all signals */
452 sigfillset(&blocked);
453 sigprocmask(SIG_BLOCK, &blocked, NULL);
454 flush_signals(current);
456 /* Become as one with the init task */
458 daemonize_fs_struct();
460 current->files = init_task.files;
461 atomic_inc(¤t->files->count);
463 reparent_to_kthreadd();
466 EXPORT_SYMBOL(daemonize);
468 static void close_files(struct files_struct * files)
476 * It is safe to dereference the fd table without RCU or
477 * ->file_lock because this is the last reference to the
478 * files structure. But use RCU to shut RCU-lockdep up.
481 fdt = files_fdtable(files);
486 if (i >= fdt->max_fds)
488 set = fdt->open_fds->fds_bits[j++];
491 struct file * file = xchg(&fdt->fd[i], NULL);
493 filp_close(file, files);
503 struct files_struct *get_files_struct(struct task_struct *task)
505 struct files_struct *files;
510 atomic_inc(&files->count);
516 void put_files_struct(struct files_struct *files)
520 if (atomic_dec_and_test(&files->count)) {
523 * Free the fd and fdset arrays if we expanded them.
524 * If the fdtable was embedded, pass files for freeing
525 * at the end of the RCU grace period. Otherwise,
526 * you can free files immediately.
529 fdt = files_fdtable(files);
530 if (fdt != &files->fdtab)
531 kmem_cache_free(files_cachep, files);
537 void reset_files_struct(struct files_struct *files)
539 struct task_struct *tsk = current;
540 struct files_struct *old;
546 put_files_struct(old);
549 void exit_files(struct task_struct *tsk)
551 struct files_struct * files = tsk->files;
557 put_files_struct(files);
561 #ifdef CONFIG_MM_OWNER
563 * A task is exiting. If it owned this mm, find a new owner for the mm.
565 void mm_update_next_owner(struct mm_struct *mm)
567 struct task_struct *c, *g, *p = current;
571 * If the exiting or execing task is not the owner, it's
572 * someone else's problem.
577 * The current owner is exiting/execing and there are no other
578 * candidates. Do not leave the mm pointing to a possibly
579 * freed task structure.
581 if (atomic_read(&mm->mm_users) <= 1) {
586 read_lock(&tasklist_lock);
588 * Search in the children
590 list_for_each_entry(c, &p->children, sibling) {
592 goto assign_new_owner;
596 * Search in the siblings
598 list_for_each_entry(c, &p->real_parent->children, sibling) {
600 goto assign_new_owner;
604 * Search through everything else. We should not get
607 do_each_thread(g, c) {
609 goto assign_new_owner;
610 } while_each_thread(g, c);
612 read_unlock(&tasklist_lock);
614 * We found no owner yet mm_users > 1: this implies that we are
615 * most likely racing with swapoff (try_to_unuse()) or /proc or
616 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
625 * The task_lock protects c->mm from changing.
626 * We always want mm->owner->mm == mm
630 * Delay read_unlock() till we have the task_lock()
631 * to ensure that c does not slip away underneath us
633 read_unlock(&tasklist_lock);
643 #endif /* CONFIG_MM_OWNER */
646 * Turn us into a lazy TLB process if we
649 static void exit_mm(struct task_struct * tsk)
651 struct mm_struct *mm = tsk->mm;
652 struct core_state *core_state;
658 * Serialize with any possible pending coredump.
659 * We must hold mmap_sem around checking core_state
660 * and clearing tsk->mm. The core-inducing thread
661 * will increment ->nr_threads for each thread in the
662 * group with ->mm != NULL.
664 down_read(&mm->mmap_sem);
665 core_state = mm->core_state;
667 struct core_thread self;
668 up_read(&mm->mmap_sem);
671 self.next = xchg(&core_state->dumper.next, &self);
673 * Implies mb(), the result of xchg() must be visible
674 * to core_state->dumper.
676 if (atomic_dec_and_test(&core_state->nr_threads))
677 complete(&core_state->startup);
680 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
681 if (!self.task) /* see coredump_finish() */
685 __set_task_state(tsk, TASK_RUNNING);
686 down_read(&mm->mmap_sem);
688 atomic_inc(&mm->mm_count);
689 BUG_ON(mm != tsk->active_mm);
690 /* more a memory barrier than a real lock */
693 up_read(&mm->mmap_sem);
694 enter_lazy_tlb(mm, current);
695 /* We don't want this task to be frozen prematurely */
696 clear_freeze_flag(tsk);
697 if (tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
698 atomic_dec(&mm->oom_disable_count);
700 mm_update_next_owner(mm);
705 * When we die, we re-parent all our children.
706 * Try to give them to another thread in our thread
707 * group, and if no such member exists, give it to
708 * the child reaper process (ie "init") in our pid
711 static struct task_struct *find_new_reaper(struct task_struct *father)
712 __releases(&tasklist_lock)
713 __acquires(&tasklist_lock)
715 struct pid_namespace *pid_ns = task_active_pid_ns(father);
716 struct task_struct *thread;
719 while_each_thread(father, thread) {
720 if (thread->flags & PF_EXITING)
722 if (unlikely(pid_ns->child_reaper == father))
723 pid_ns->child_reaper = thread;
727 if (unlikely(pid_ns->child_reaper == father)) {
728 write_unlock_irq(&tasklist_lock);
729 if (unlikely(pid_ns == &init_pid_ns))
730 panic("Attempted to kill init!");
732 zap_pid_ns_processes(pid_ns);
733 write_lock_irq(&tasklist_lock);
735 * We can not clear ->child_reaper or leave it alone.
736 * There may by stealth EXIT_DEAD tasks on ->children,
737 * forget_original_parent() must move them somewhere.
739 pid_ns->child_reaper = init_pid_ns.child_reaper;
742 return pid_ns->child_reaper;
746 * Any that need to be release_task'd are put on the @dead list.
748 static void reparent_leader(struct task_struct *father, struct task_struct *p,
749 struct list_head *dead)
751 list_move_tail(&p->sibling, &p->real_parent->children);
753 if (task_detached(p))
756 * If this is a threaded reparent there is no need to
757 * notify anyone anything has happened.
759 if (same_thread_group(p->real_parent, father))
762 /* We don't want people slaying init. */
763 p->exit_signal = SIGCHLD;
765 /* If it has exited notify the new parent about this child's death. */
766 if (!task_ptrace(p) &&
767 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
768 do_notify_parent(p, p->exit_signal);
769 if (task_detached(p)) {
770 p->exit_state = EXIT_DEAD;
771 list_move_tail(&p->sibling, dead);
775 kill_orphaned_pgrp(p, father);
778 static void forget_original_parent(struct task_struct *father)
780 struct task_struct *p, *n, *reaper;
781 LIST_HEAD(dead_children);
783 write_lock_irq(&tasklist_lock);
785 * Note that exit_ptrace() and find_new_reaper() might
786 * drop tasklist_lock and reacquire it.
789 reaper = find_new_reaper(father);
791 list_for_each_entry_safe(p, n, &father->children, sibling) {
792 struct task_struct *t = p;
794 t->real_parent = reaper;
795 if (t->parent == father) {
796 BUG_ON(task_ptrace(t));
797 t->parent = t->real_parent;
799 if (t->pdeath_signal)
800 group_send_sig_info(t->pdeath_signal,
802 } while_each_thread(p, t);
803 reparent_leader(father, p, &dead_children);
805 write_unlock_irq(&tasklist_lock);
807 BUG_ON(!list_empty(&father->children));
809 list_for_each_entry_safe(p, n, &dead_children, sibling) {
810 list_del_init(&p->sibling);
816 * Send signals to all our closest relatives so that they know
817 * to properly mourn us..
819 static void exit_notify(struct task_struct *tsk, int group_dead)
825 * This does two things:
827 * A. Make init inherit all the child processes
828 * B. Check to see if any process groups have become orphaned
829 * as a result of our exiting, and if they have any stopped
830 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
832 forget_original_parent(tsk);
833 exit_task_namespaces(tsk);
835 write_lock_irq(&tasklist_lock);
837 kill_orphaned_pgrp(tsk->group_leader, NULL);
839 /* Let father know we died
841 * Thread signals are configurable, but you aren't going to use
842 * that to send signals to arbitrary processes.
843 * That stops right now.
845 * If the parent exec id doesn't match the exec id we saved
846 * when we started then we know the parent has changed security
849 * If our self_exec id doesn't match our parent_exec_id then
850 * we have changed execution domain as these two values started
851 * the same after a fork.
853 if (tsk->exit_signal != SIGCHLD && !task_detached(tsk) &&
854 (tsk->parent_exec_id != tsk->real_parent->self_exec_id ||
855 tsk->self_exec_id != tsk->parent_exec_id))
856 tsk->exit_signal = SIGCHLD;
858 signal = tracehook_notify_death(tsk, &cookie, group_dead);
860 signal = do_notify_parent(tsk, signal);
862 tsk->exit_state = signal == DEATH_REAP ? EXIT_DEAD : EXIT_ZOMBIE;
864 /* mt-exec, de_thread() is waiting for group leader */
865 if (unlikely(tsk->signal->notify_count < 0))
866 wake_up_process(tsk->signal->group_exit_task);
867 write_unlock_irq(&tasklist_lock);
869 tracehook_report_death(tsk, signal, cookie, group_dead);
871 /* If the process is dead, release it - nobody will wait for it */
872 if (signal == DEATH_REAP)
876 #ifdef CONFIG_DEBUG_STACK_USAGE
877 static void check_stack_usage(void)
879 static DEFINE_SPINLOCK(low_water_lock);
880 static int lowest_to_date = THREAD_SIZE;
883 free = stack_not_used(current);
885 if (free >= lowest_to_date)
888 spin_lock(&low_water_lock);
889 if (free < lowest_to_date) {
890 printk(KERN_WARNING "%s used greatest stack depth: %lu bytes "
892 current->comm, free);
893 lowest_to_date = free;
895 spin_unlock(&low_water_lock);
898 static inline void check_stack_usage(void) {}
901 NORET_TYPE void do_exit(long code)
903 struct task_struct *tsk = current;
906 profile_task_exit(tsk);
908 WARN_ON(atomic_read(&tsk->fs_excl));
909 WARN_ON(blk_needs_flush_plug(tsk));
911 if (unlikely(in_interrupt()))
912 panic("Aiee, killing interrupt handler!");
913 if (unlikely(!tsk->pid))
914 panic("Attempted to kill the idle task!");
917 * If do_exit is called because this processes oopsed, it's possible
918 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
919 * continuing. Amongst other possible reasons, this is to prevent
920 * mm_release()->clear_child_tid() from writing to a user-controlled
925 tracehook_report_exit(&code);
927 validate_creds_for_do_exit(tsk);
930 * We're taking recursive faults here in do_exit. Safest is to just
931 * leave this task alone and wait for reboot.
933 if (unlikely(tsk->flags & PF_EXITING)) {
935 "Fixing recursive fault but reboot is needed!\n");
937 * We can do this unlocked here. The futex code uses
938 * this flag just to verify whether the pi state
939 * cleanup has been done or not. In the worst case it
940 * loops once more. We pretend that the cleanup was
941 * done as there is no way to return. Either the
942 * OWNER_DIED bit is set by now or we push the blocked
943 * task into the wait for ever nirwana as well.
945 tsk->flags |= PF_EXITPIDONE;
946 set_current_state(TASK_UNINTERRUPTIBLE);
952 exit_signals(tsk); /* sets PF_EXITING */
954 * tsk->flags are checked in the futex code to protect against
955 * an exiting task cleaning up the robust pi futexes.
958 raw_spin_unlock_wait(&tsk->pi_lock);
960 if (unlikely(in_atomic()))
961 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
962 current->comm, task_pid_nr(current),
965 acct_update_integrals(tsk);
966 /* sync mm's RSS info before statistics gathering */
968 sync_mm_rss(tsk, tsk->mm);
969 group_dead = atomic_dec_and_test(&tsk->signal->live);
971 hrtimer_cancel(&tsk->signal->real_timer);
972 exit_itimers(tsk->signal);
974 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
976 acct_collect(code, group_dead);
979 if (unlikely(tsk->audit_context))
982 tsk->exit_code = code;
983 taskstats_exit(tsk, group_dead);
989 trace_sched_process_exit(tsk);
998 * Flush inherited counters to the parent - before the parent
999 * gets woken up by child-exit notifications.
1001 * because of cgroup mode, must be called before cgroup_exit()
1003 perf_event_exit_task(tsk);
1005 cgroup_exit(tsk, 1);
1008 disassociate_ctty(1);
1010 module_put(task_thread_info(tsk)->exec_domain->module);
1012 proc_exit_connector(tsk);
1015 * FIXME: do that only when needed, using sched_exit tracepoint
1017 ptrace_put_breakpoints(tsk);
1019 exit_notify(tsk, group_dead);
1022 mpol_put(tsk->mempolicy);
1023 tsk->mempolicy = NULL;
1027 if (unlikely(current->pi_state_cache))
1028 kfree(current->pi_state_cache);
1031 * Make sure we are holding no locks:
1033 debug_check_no_locks_held(tsk);
1035 * We can do this unlocked here. The futex code uses this flag
1036 * just to verify whether the pi state cleanup has been done
1037 * or not. In the worst case it loops once more.
1039 tsk->flags |= PF_EXITPIDONE;
1041 if (tsk->io_context)
1042 exit_io_context(tsk);
1044 if (tsk->splice_pipe)
1045 __free_pipe_info(tsk->splice_pipe);
1047 validate_creds_for_do_exit(tsk);
1051 /* causes final put_task_struct in finish_task_switch(). */
1052 tsk->state = TASK_DEAD;
1055 /* Avoid "noreturn function does return". */
1057 cpu_relax(); /* For when BUG is null */
1060 EXPORT_SYMBOL_GPL(do_exit);
1062 NORET_TYPE void complete_and_exit(struct completion *comp, long code)
1070 EXPORT_SYMBOL(complete_and_exit);
1072 SYSCALL_DEFINE1(exit, int, error_code)
1074 do_exit((error_code&0xff)<<8);
1078 * Take down every thread in the group. This is called by fatal signals
1079 * as well as by sys_exit_group (below).
1082 do_group_exit(int exit_code)
1084 struct signal_struct *sig = current->signal;
1086 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
1088 if (signal_group_exit(sig))
1089 exit_code = sig->group_exit_code;
1090 else if (!thread_group_empty(current)) {
1091 struct sighand_struct *const sighand = current->sighand;
1092 spin_lock_irq(&sighand->siglock);
1093 if (signal_group_exit(sig))
1094 /* Another thread got here before we took the lock. */
1095 exit_code = sig->group_exit_code;
1097 sig->group_exit_code = exit_code;
1098 sig->flags = SIGNAL_GROUP_EXIT;
1099 zap_other_threads(current);
1101 spin_unlock_irq(&sighand->siglock);
1109 * this kills every thread in the thread group. Note that any externally
1110 * wait4()-ing process will get the correct exit code - even if this
1111 * thread is not the thread group leader.
1113 SYSCALL_DEFINE1(exit_group, int, error_code)
1115 do_group_exit((error_code & 0xff) << 8);
1121 enum pid_type wo_type;
1125 struct siginfo __user *wo_info;
1126 int __user *wo_stat;
1127 struct rusage __user *wo_rusage;
1129 wait_queue_t child_wait;
1134 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
1136 if (type != PIDTYPE_PID)
1137 task = task->group_leader;
1138 return task->pids[type].pid;
1141 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
1143 return wo->wo_type == PIDTYPE_MAX ||
1144 task_pid_type(p, wo->wo_type) == wo->wo_pid;
1147 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
1149 if (!eligible_pid(wo, p))
1151 /* Wait for all children (clone and not) if __WALL is set;
1152 * otherwise, wait for clone children *only* if __WCLONE is
1153 * set; otherwise, wait for non-clone children *only*. (Note:
1154 * A "clone" child here is one that reports to its parent
1155 * using a signal other than SIGCHLD.) */
1156 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
1157 && !(wo->wo_flags & __WALL))
1163 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
1164 pid_t pid, uid_t uid, int why, int status)
1166 struct siginfo __user *infop;
1167 int retval = wo->wo_rusage
1168 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1171 infop = wo->wo_info;
1174 retval = put_user(SIGCHLD, &infop->si_signo);
1176 retval = put_user(0, &infop->si_errno);
1178 retval = put_user((short)why, &infop->si_code);
1180 retval = put_user(pid, &infop->si_pid);
1182 retval = put_user(uid, &infop->si_uid);
1184 retval = put_user(status, &infop->si_status);
1192 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1193 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1194 * the lock and this task is uninteresting. If we return nonzero, we have
1195 * released the lock and the system call should return.
1197 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1199 unsigned long state;
1200 int retval, status, traced;
1201 pid_t pid = task_pid_vnr(p);
1202 uid_t uid = __task_cred(p)->uid;
1203 struct siginfo __user *infop;
1205 if (!likely(wo->wo_flags & WEXITED))
1208 if (unlikely(wo->wo_flags & WNOWAIT)) {
1209 int exit_code = p->exit_code;
1213 read_unlock(&tasklist_lock);
1214 if ((exit_code & 0x7f) == 0) {
1216 status = exit_code >> 8;
1218 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1219 status = exit_code & 0x7f;
1221 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1225 * Try to move the task's state to DEAD
1226 * only one thread is allowed to do this:
1228 state = xchg(&p->exit_state, EXIT_DEAD);
1229 if (state != EXIT_ZOMBIE) {
1230 BUG_ON(state != EXIT_DEAD);
1234 traced = ptrace_reparented(p);
1236 * It can be ptraced but not reparented, check
1237 * !task_detached() to filter out sub-threads.
1239 if (likely(!traced) && likely(!task_detached(p))) {
1240 struct signal_struct *psig;
1241 struct signal_struct *sig;
1242 unsigned long maxrss;
1243 cputime_t tgutime, tgstime;
1246 * The resource counters for the group leader are in its
1247 * own task_struct. Those for dead threads in the group
1248 * are in its signal_struct, as are those for the child
1249 * processes it has previously reaped. All these
1250 * accumulate in the parent's signal_struct c* fields.
1252 * We don't bother to take a lock here to protect these
1253 * p->signal fields, because they are only touched by
1254 * __exit_signal, which runs with tasklist_lock
1255 * write-locked anyway, and so is excluded here. We do
1256 * need to protect the access to parent->signal fields,
1257 * as other threads in the parent group can be right
1258 * here reaping other children at the same time.
1260 * We use thread_group_times() to get times for the thread
1261 * group, which consolidates times for all threads in the
1262 * group including the group leader.
1264 thread_group_times(p, &tgutime, &tgstime);
1265 spin_lock_irq(&p->real_parent->sighand->siglock);
1266 psig = p->real_parent->signal;
1269 cputime_add(psig->cutime,
1270 cputime_add(tgutime,
1273 cputime_add(psig->cstime,
1274 cputime_add(tgstime,
1277 cputime_add(psig->cgtime,
1278 cputime_add(p->gtime,
1279 cputime_add(sig->gtime,
1282 p->min_flt + sig->min_flt + sig->cmin_flt;
1284 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1286 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1288 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1290 task_io_get_inblock(p) +
1291 sig->inblock + sig->cinblock;
1293 task_io_get_oublock(p) +
1294 sig->oublock + sig->coublock;
1295 maxrss = max(sig->maxrss, sig->cmaxrss);
1296 if (psig->cmaxrss < maxrss)
1297 psig->cmaxrss = maxrss;
1298 task_io_accounting_add(&psig->ioac, &p->ioac);
1299 task_io_accounting_add(&psig->ioac, &sig->ioac);
1300 spin_unlock_irq(&p->real_parent->sighand->siglock);
1304 * Now we are sure this task is interesting, and no other
1305 * thread can reap it because we set its state to EXIT_DEAD.
1307 read_unlock(&tasklist_lock);
1309 retval = wo->wo_rusage
1310 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1311 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1312 ? p->signal->group_exit_code : p->exit_code;
1313 if (!retval && wo->wo_stat)
1314 retval = put_user(status, wo->wo_stat);
1316 infop = wo->wo_info;
1317 if (!retval && infop)
1318 retval = put_user(SIGCHLD, &infop->si_signo);
1319 if (!retval && infop)
1320 retval = put_user(0, &infop->si_errno);
1321 if (!retval && infop) {
1324 if ((status & 0x7f) == 0) {
1328 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1331 retval = put_user((short)why, &infop->si_code);
1333 retval = put_user(status, &infop->si_status);
1335 if (!retval && infop)
1336 retval = put_user(pid, &infop->si_pid);
1337 if (!retval && infop)
1338 retval = put_user(uid, &infop->si_uid);
1343 write_lock_irq(&tasklist_lock);
1344 /* We dropped tasklist, ptracer could die and untrace */
1347 * If this is not a detached task, notify the parent.
1348 * If it's still not detached after that, don't release
1351 if (!task_detached(p)) {
1352 do_notify_parent(p, p->exit_signal);
1353 if (!task_detached(p)) {
1354 p->exit_state = EXIT_ZOMBIE;
1358 write_unlock_irq(&tasklist_lock);
1366 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1369 if (task_is_stopped_or_traced(p))
1370 return &p->exit_code;
1372 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1373 return &p->signal->group_exit_code;
1379 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1381 * @ptrace: is the wait for ptrace
1382 * @p: task to wait for
1384 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1387 * read_lock(&tasklist_lock), which is released if return value is
1388 * non-zero. Also, grabs and releases @p->sighand->siglock.
1391 * 0 if wait condition didn't exist and search for other wait conditions
1392 * should continue. Non-zero return, -errno on failure and @p's pid on
1393 * success, implies that tasklist_lock is released and wait condition
1394 * search should terminate.
1396 static int wait_task_stopped(struct wait_opts *wo,
1397 int ptrace, struct task_struct *p)
1399 struct siginfo __user *infop;
1400 int retval, exit_code, *p_code, why;
1401 uid_t uid = 0; /* unneeded, required by compiler */
1405 * Traditionally we see ptrace'd stopped tasks regardless of options.
1407 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1410 if (!task_stopped_code(p, ptrace))
1414 spin_lock_irq(&p->sighand->siglock);
1416 p_code = task_stopped_code(p, ptrace);
1417 if (unlikely(!p_code))
1420 exit_code = *p_code;
1424 if (!unlikely(wo->wo_flags & WNOWAIT))
1429 spin_unlock_irq(&p->sighand->siglock);
1434 * Now we are pretty sure this task is interesting.
1435 * Make sure it doesn't get reaped out from under us while we
1436 * give up the lock and then examine it below. We don't want to
1437 * keep holding onto the tasklist_lock while we call getrusage and
1438 * possibly take page faults for user memory.
1441 pid = task_pid_vnr(p);
1442 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1443 read_unlock(&tasklist_lock);
1445 if (unlikely(wo->wo_flags & WNOWAIT))
1446 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1448 retval = wo->wo_rusage
1449 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1450 if (!retval && wo->wo_stat)
1451 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1453 infop = wo->wo_info;
1454 if (!retval && infop)
1455 retval = put_user(SIGCHLD, &infop->si_signo);
1456 if (!retval && infop)
1457 retval = put_user(0, &infop->si_errno);
1458 if (!retval && infop)
1459 retval = put_user((short)why, &infop->si_code);
1460 if (!retval && infop)
1461 retval = put_user(exit_code, &infop->si_status);
1462 if (!retval && infop)
1463 retval = put_user(pid, &infop->si_pid);
1464 if (!retval && infop)
1465 retval = put_user(uid, &infop->si_uid);
1475 * Handle do_wait work for one task in a live, non-stopped state.
1476 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1477 * the lock and this task is uninteresting. If we return nonzero, we have
1478 * released the lock and the system call should return.
1480 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1486 if (!unlikely(wo->wo_flags & WCONTINUED))
1489 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1492 spin_lock_irq(&p->sighand->siglock);
1493 /* Re-check with the lock held. */
1494 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1495 spin_unlock_irq(&p->sighand->siglock);
1498 if (!unlikely(wo->wo_flags & WNOWAIT))
1499 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1501 spin_unlock_irq(&p->sighand->siglock);
1503 pid = task_pid_vnr(p);
1505 read_unlock(&tasklist_lock);
1508 retval = wo->wo_rusage
1509 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1511 if (!retval && wo->wo_stat)
1512 retval = put_user(0xffff, wo->wo_stat);
1516 retval = wait_noreap_copyout(wo, p, pid, uid,
1517 CLD_CONTINUED, SIGCONT);
1518 BUG_ON(retval == 0);
1525 * Consider @p for a wait by @parent.
1527 * -ECHILD should be in ->notask_error before the first call.
1528 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1529 * Returns zero if the search for a child should continue;
1530 * then ->notask_error is 0 if @p is an eligible child,
1531 * or another error from security_task_wait(), or still -ECHILD.
1533 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1534 struct task_struct *p)
1536 int ret = eligible_child(wo, p);
1540 ret = security_task_wait(p);
1541 if (unlikely(ret < 0)) {
1543 * If we have not yet seen any eligible child,
1544 * then let this error code replace -ECHILD.
1545 * A permission error will give the user a clue
1546 * to look for security policy problems, rather
1547 * than for mysterious wait bugs.
1549 if (wo->notask_error)
1550 wo->notask_error = ret;
1554 /* dead body doesn't have much to contribute */
1555 if (p->exit_state == EXIT_DEAD)
1559 if (p->exit_state == EXIT_ZOMBIE) {
1561 * A zombie ptracee is only visible to its ptracer.
1562 * Notification and reaping will be cascaded to the real
1563 * parent when the ptracer detaches.
1565 if (likely(!ptrace) && unlikely(task_ptrace(p))) {
1566 /* it will become visible, clear notask_error */
1567 wo->notask_error = 0;
1571 /* we don't reap group leaders with subthreads */
1572 if (!delay_group_leader(p))
1573 return wait_task_zombie(wo, p);
1576 * Allow access to stopped/continued state via zombie by
1577 * falling through. Clearing of notask_error is complex.
1581 * If WEXITED is set, notask_error should naturally be
1582 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1583 * so, if there are live subthreads, there are events to
1584 * wait for. If all subthreads are dead, it's still safe
1585 * to clear - this function will be called again in finite
1586 * amount time once all the subthreads are released and
1587 * will then return without clearing.
1591 * Stopped state is per-task and thus can't change once the
1592 * target task dies. Only continued and exited can happen.
1593 * Clear notask_error if WCONTINUED | WEXITED.
1595 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1596 wo->notask_error = 0;
1599 * If @p is ptraced by a task in its real parent's group,
1600 * hide group stop/continued state when looking at @p as
1601 * the real parent; otherwise, a single stop can be
1602 * reported twice as group and ptrace stops.
1604 * If a ptracer wants to distinguish the two events for its
1605 * own children, it should create a separate process which
1606 * takes the role of real parent.
1608 if (likely(!ptrace) && task_ptrace(p) &&
1609 same_thread_group(p->parent, p->real_parent))
1613 * @p is alive and it's gonna stop, continue or exit, so
1614 * there always is something to wait for.
1616 wo->notask_error = 0;
1620 * Wait for stopped. Depending on @ptrace, different stopped state
1621 * is used and the two don't interact with each other.
1623 ret = wait_task_stopped(wo, ptrace, p);
1628 * Wait for continued. There's only one continued state and the
1629 * ptracer can consume it which can confuse the real parent. Don't
1630 * use WCONTINUED from ptracer. You don't need or want it.
1632 return wait_task_continued(wo, p);
1636 * Do the work of do_wait() for one thread in the group, @tsk.
1638 * -ECHILD should be in ->notask_error before the first call.
1639 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1640 * Returns zero if the search for a child should continue; then
1641 * ->notask_error is 0 if there were any eligible children,
1642 * or another error from security_task_wait(), or still -ECHILD.
1644 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1646 struct task_struct *p;
1648 list_for_each_entry(p, &tsk->children, sibling) {
1649 int ret = wait_consider_task(wo, 0, p);
1657 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1659 struct task_struct *p;
1661 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1662 int ret = wait_consider_task(wo, 1, p);
1670 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1671 int sync, void *key)
1673 struct wait_opts *wo = container_of(wait, struct wait_opts,
1675 struct task_struct *p = key;
1677 if (!eligible_pid(wo, p))
1680 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1683 return default_wake_function(wait, mode, sync, key);
1686 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1688 __wake_up_sync_key(&parent->signal->wait_chldexit,
1689 TASK_INTERRUPTIBLE, 1, p);
1692 static long do_wait(struct wait_opts *wo)
1694 struct task_struct *tsk;
1697 trace_sched_process_wait(wo->wo_pid);
1699 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1700 wo->child_wait.private = current;
1701 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1704 * If there is nothing that can match our critiera just get out.
1705 * We will clear ->notask_error to zero if we see any child that
1706 * might later match our criteria, even if we are not able to reap
1709 wo->notask_error = -ECHILD;
1710 if ((wo->wo_type < PIDTYPE_MAX) &&
1711 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1714 set_current_state(TASK_INTERRUPTIBLE);
1715 read_lock(&tasklist_lock);
1718 retval = do_wait_thread(wo, tsk);
1722 retval = ptrace_do_wait(wo, tsk);
1726 if (wo->wo_flags & __WNOTHREAD)
1728 } while_each_thread(current, tsk);
1729 read_unlock(&tasklist_lock);
1732 retval = wo->notask_error;
1733 if (!retval && !(wo->wo_flags & WNOHANG)) {
1734 retval = -ERESTARTSYS;
1735 if (!signal_pending(current)) {
1741 __set_current_state(TASK_RUNNING);
1742 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1746 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1747 infop, int, options, struct rusage __user *, ru)
1749 struct wait_opts wo;
1750 struct pid *pid = NULL;
1754 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1756 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1769 type = PIDTYPE_PGID;
1777 if (type < PIDTYPE_MAX)
1778 pid = find_get_pid(upid);
1782 wo.wo_flags = options;
1792 * For a WNOHANG return, clear out all the fields
1793 * we would set so the user can easily tell the
1797 ret = put_user(0, &infop->si_signo);
1799 ret = put_user(0, &infop->si_errno);
1801 ret = put_user(0, &infop->si_code);
1803 ret = put_user(0, &infop->si_pid);
1805 ret = put_user(0, &infop->si_uid);
1807 ret = put_user(0, &infop->si_status);
1812 /* avoid REGPARM breakage on x86: */
1813 asmlinkage_protect(5, ret, which, upid, infop, options, ru);
1817 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1818 int, options, struct rusage __user *, ru)
1820 struct wait_opts wo;
1821 struct pid *pid = NULL;
1825 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1826 __WNOTHREAD|__WCLONE|__WALL))
1831 else if (upid < 0) {
1832 type = PIDTYPE_PGID;
1833 pid = find_get_pid(-upid);
1834 } else if (upid == 0) {
1835 type = PIDTYPE_PGID;
1836 pid = get_task_pid(current, PIDTYPE_PGID);
1837 } else /* upid > 0 */ {
1839 pid = find_get_pid(upid);
1844 wo.wo_flags = options | WEXITED;
1846 wo.wo_stat = stat_addr;
1851 /* avoid REGPARM breakage on x86: */
1852 asmlinkage_protect(4, ret, upid, stat_addr, options, ru);
1856 #ifdef __ARCH_WANT_SYS_WAITPID
1859 * sys_waitpid() remains for compatibility. waitpid() should be
1860 * implemented by calling sys_wait4() from libc.a.
1862 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1864 return sys_wait4(pid, stat_addr, options, NULL);