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/freezer.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/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>
54 #include <linux/writeback.h>
55 #include <linux/shm.h>
57 #include <asm/uaccess.h>
58 #include <asm/unistd.h>
59 #include <asm/pgtable.h>
60 #include <asm/mmu_context.h>
62 static void exit_mm(struct task_struct * tsk);
64 static void __unhash_process(struct task_struct *p, bool group_dead)
67 detach_pid(p, PIDTYPE_PID);
69 detach_pid(p, PIDTYPE_PGID);
70 detach_pid(p, PIDTYPE_SID);
72 list_del_rcu(&p->tasks);
73 list_del_init(&p->sibling);
74 __this_cpu_dec(process_counts);
76 list_del_rcu(&p->thread_group);
80 * This function expects the tasklist_lock write-locked.
82 static void __exit_signal(struct task_struct *tsk)
84 struct signal_struct *sig = tsk->signal;
85 bool group_dead = thread_group_leader(tsk);
86 struct sighand_struct *sighand;
87 struct tty_struct *uninitialized_var(tty);
88 cputime_t utime, stime;
90 sighand = rcu_dereference_check(tsk->sighand,
91 lockdep_tasklist_lock_is_held());
92 spin_lock(&sighand->siglock);
94 posix_cpu_timers_exit(tsk);
96 posix_cpu_timers_exit_group(tsk);
101 * This can only happen if the caller is de_thread().
102 * FIXME: this is the temporary hack, we should teach
103 * posix-cpu-timers to handle this case correctly.
105 if (unlikely(has_group_leader_pid(tsk)))
106 posix_cpu_timers_exit_group(tsk);
109 * If there is any task waiting for the group exit
112 if (sig->notify_count > 0 && !--sig->notify_count)
113 wake_up_process(sig->group_exit_task);
115 if (tsk == sig->curr_target)
116 sig->curr_target = next_thread(tsk);
118 * Accumulate here the counters for all threads but the
119 * group leader as they die, so they can be added into
120 * the process-wide totals when those are taken.
121 * The group leader stays around as a zombie as long
122 * as there are other threads. When it gets reaped,
123 * the exit.c code will add its counts into these totals.
124 * We won't ever get here for the group leader, since it
125 * will have been the last reference on the signal_struct.
127 task_cputime(tsk, &utime, &stime);
130 sig->gtime += task_gtime(tsk);
131 sig->min_flt += tsk->min_flt;
132 sig->maj_flt += tsk->maj_flt;
133 sig->nvcsw += tsk->nvcsw;
134 sig->nivcsw += tsk->nivcsw;
135 sig->inblock += task_io_get_inblock(tsk);
136 sig->oublock += task_io_get_oublock(tsk);
137 task_io_accounting_add(&sig->ioac, &tsk->ioac);
138 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
142 __unhash_process(tsk, group_dead);
145 * Do this under ->siglock, we can race with another thread
146 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
148 flush_sigqueue(&tsk->pending);
150 spin_unlock(&sighand->siglock);
152 __cleanup_sighand(sighand);
153 clear_tsk_thread_flag(tsk,TIF_SIGPENDING);
155 flush_sigqueue(&sig->shared_pending);
160 static void delayed_put_task_struct(struct rcu_head *rhp)
162 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
164 perf_event_delayed_put(tsk);
165 trace_sched_process_free(tsk);
166 put_task_struct(tsk);
170 void release_task(struct task_struct * p)
172 struct task_struct *leader;
175 /* don't need to get the RCU readlock here - the process is dead and
176 * can't be modifying its own credentials. But shut RCU-lockdep up */
178 atomic_dec(&__task_cred(p)->user->processes);
183 write_lock_irq(&tasklist_lock);
184 ptrace_release_task(p);
188 * If we are the last non-leader member of the thread
189 * group, and the leader is zombie, then notify the
190 * group leader's parent process. (if it wants notification.)
193 leader = p->group_leader;
194 if (leader != p && thread_group_empty(leader) && leader->exit_state == EXIT_ZOMBIE) {
196 * If we were the last child thread and the leader has
197 * exited already, and the leader's parent ignores SIGCHLD,
198 * then we are the one who should release the leader.
200 zap_leader = do_notify_parent(leader, leader->exit_signal);
202 leader->exit_state = EXIT_DEAD;
205 write_unlock_irq(&tasklist_lock);
207 call_rcu(&p->rcu, delayed_put_task_struct);
210 if (unlikely(zap_leader))
215 * This checks not only the pgrp, but falls back on the pid if no
216 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
219 * The caller must hold rcu lock or the tasklist lock.
221 struct pid *session_of_pgrp(struct pid *pgrp)
223 struct task_struct *p;
224 struct pid *sid = NULL;
226 p = pid_task(pgrp, PIDTYPE_PGID);
228 p = pid_task(pgrp, PIDTYPE_PID);
230 sid = task_session(p);
236 * Determine if a process group is "orphaned", according to the POSIX
237 * definition in 2.2.2.52. Orphaned process groups are not to be affected
238 * by terminal-generated stop signals. Newly orphaned process groups are
239 * to receive a SIGHUP and a SIGCONT.
241 * "I ask you, have you ever known what it is to be an orphan?"
243 static int will_become_orphaned_pgrp(struct pid *pgrp, struct task_struct *ignored_task)
245 struct task_struct *p;
247 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
248 if ((p == ignored_task) ||
249 (p->exit_state && thread_group_empty(p)) ||
250 is_global_init(p->real_parent))
253 if (task_pgrp(p->real_parent) != pgrp &&
254 task_session(p->real_parent) == task_session(p))
256 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
261 int is_current_pgrp_orphaned(void)
265 read_lock(&tasklist_lock);
266 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
267 read_unlock(&tasklist_lock);
272 static bool has_stopped_jobs(struct pid *pgrp)
274 struct task_struct *p;
276 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
277 if (p->signal->flags & SIGNAL_STOP_STOPPED)
279 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
285 * Check to see if any process groups have become orphaned as
286 * a result of our exiting, and if they have any stopped jobs,
287 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
290 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
292 struct pid *pgrp = task_pgrp(tsk);
293 struct task_struct *ignored_task = tsk;
296 /* exit: our father is in a different pgrp than
297 * we are and we were the only connection outside.
299 parent = tsk->real_parent;
301 /* reparent: our child is in a different pgrp than
302 * we are, and it was the only connection outside.
306 if (task_pgrp(parent) != pgrp &&
307 task_session(parent) == task_session(tsk) &&
308 will_become_orphaned_pgrp(pgrp, ignored_task) &&
309 has_stopped_jobs(pgrp)) {
310 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
311 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
315 void __set_special_pids(struct pid *pid)
317 struct task_struct *curr = current->group_leader;
319 if (task_session(curr) != pid)
320 change_pid(curr, PIDTYPE_SID, pid);
322 if (task_pgrp(curr) != pid)
323 change_pid(curr, PIDTYPE_PGID, pid);
327 * Let kernel threads use this to say that they allow a certain signal.
328 * Must not be used if kthread was cloned with CLONE_SIGHAND.
330 int allow_signal(int sig)
332 if (!valid_signal(sig) || sig < 1)
335 spin_lock_irq(¤t->sighand->siglock);
336 /* This is only needed for daemonize()'ed kthreads */
337 sigdelset(¤t->blocked, sig);
339 * Kernel threads handle their own signals. Let the signal code
340 * know it'll be handled, so that they don't get converted to
341 * SIGKILL or just silently dropped.
343 current->sighand->action[(sig)-1].sa.sa_handler = (void __user *)2;
345 spin_unlock_irq(¤t->sighand->siglock);
349 EXPORT_SYMBOL(allow_signal);
351 int disallow_signal(int sig)
353 if (!valid_signal(sig) || sig < 1)
356 spin_lock_irq(¤t->sighand->siglock);
357 current->sighand->action[(sig)-1].sa.sa_handler = SIG_IGN;
359 spin_unlock_irq(¤t->sighand->siglock);
363 EXPORT_SYMBOL(disallow_signal);
365 #ifdef CONFIG_MM_OWNER
367 * A task is exiting. If it owned this mm, find a new owner for the mm.
369 void mm_update_next_owner(struct mm_struct *mm)
371 struct task_struct *c, *g, *p = current;
375 * If the exiting or execing task is not the owner, it's
376 * someone else's problem.
381 * The current owner is exiting/execing and there are no other
382 * candidates. Do not leave the mm pointing to a possibly
383 * freed task structure.
385 if (atomic_read(&mm->mm_users) <= 1) {
390 read_lock(&tasklist_lock);
392 * Search in the children
394 list_for_each_entry(c, &p->children, sibling) {
396 goto assign_new_owner;
400 * Search in the siblings
402 list_for_each_entry(c, &p->real_parent->children, sibling) {
404 goto assign_new_owner;
408 * Search through everything else. We should not get
411 do_each_thread(g, c) {
413 goto assign_new_owner;
414 } while_each_thread(g, c);
416 read_unlock(&tasklist_lock);
418 * We found no owner yet mm_users > 1: this implies that we are
419 * most likely racing with swapoff (try_to_unuse()) or /proc or
420 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
429 * The task_lock protects c->mm from changing.
430 * We always want mm->owner->mm == mm
434 * Delay read_unlock() till we have the task_lock()
435 * to ensure that c does not slip away underneath us
437 read_unlock(&tasklist_lock);
447 #endif /* CONFIG_MM_OWNER */
450 * Turn us into a lazy TLB process if we
453 static void exit_mm(struct task_struct * tsk)
455 struct mm_struct *mm = tsk->mm;
456 struct core_state *core_state;
463 * Serialize with any possible pending coredump.
464 * We must hold mmap_sem around checking core_state
465 * and clearing tsk->mm. The core-inducing thread
466 * will increment ->nr_threads for each thread in the
467 * group with ->mm != NULL.
469 down_read(&mm->mmap_sem);
470 core_state = mm->core_state;
472 struct core_thread self;
473 up_read(&mm->mmap_sem);
476 self.next = xchg(&core_state->dumper.next, &self);
478 * Implies mb(), the result of xchg() must be visible
479 * to core_state->dumper.
481 if (atomic_dec_and_test(&core_state->nr_threads))
482 complete(&core_state->startup);
485 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
486 if (!self.task) /* see coredump_finish() */
488 freezable_schedule();
490 __set_task_state(tsk, TASK_RUNNING);
491 down_read(&mm->mmap_sem);
493 atomic_inc(&mm->mm_count);
494 BUG_ON(mm != tsk->active_mm);
495 /* more a memory barrier than a real lock */
498 up_read(&mm->mmap_sem);
499 enter_lazy_tlb(mm, current);
501 mm_update_next_owner(mm);
506 * When we die, we re-parent all our children, and try to:
507 * 1. give them to another thread in our thread group, if such a member exists
508 * 2. give it to the first ancestor process which prctl'd itself as a
509 * child_subreaper for its children (like a service manager)
510 * 3. give it to the init process (PID 1) in our pid namespace
512 static struct task_struct *find_new_reaper(struct task_struct *father)
513 __releases(&tasklist_lock)
514 __acquires(&tasklist_lock)
516 struct pid_namespace *pid_ns = task_active_pid_ns(father);
517 struct task_struct *thread;
520 while_each_thread(father, thread) {
521 if (thread->flags & PF_EXITING)
523 if (unlikely(pid_ns->child_reaper == father))
524 pid_ns->child_reaper = thread;
528 if (unlikely(pid_ns->child_reaper == father)) {
529 write_unlock_irq(&tasklist_lock);
530 if (unlikely(pid_ns == &init_pid_ns)) {
531 panic("Attempted to kill init! exitcode=0x%08x\n",
532 father->signal->group_exit_code ?:
536 zap_pid_ns_processes(pid_ns);
537 write_lock_irq(&tasklist_lock);
538 } else if (father->signal->has_child_subreaper) {
539 struct task_struct *reaper;
542 * Find the first ancestor marked as child_subreaper.
543 * Note that the code below checks same_thread_group(reaper,
544 * pid_ns->child_reaper). This is what we need to DTRT in a
545 * PID namespace. However we still need the check above, see
546 * http://marc.info/?l=linux-kernel&m=131385460420380
548 for (reaper = father->real_parent;
549 reaper != &init_task;
550 reaper = reaper->real_parent) {
551 if (same_thread_group(reaper, pid_ns->child_reaper))
553 if (!reaper->signal->is_child_subreaper)
557 if (!(thread->flags & PF_EXITING))
559 } while_each_thread(reaper, thread);
563 return pid_ns->child_reaper;
567 * Any that need to be release_task'd are put on the @dead list.
569 static void reparent_leader(struct task_struct *father, struct task_struct *p,
570 struct list_head *dead)
572 list_move_tail(&p->sibling, &p->real_parent->children);
574 * If this is a threaded reparent there is no need to
575 * notify anyone anything has happened.
577 if (same_thread_group(p->real_parent, father))
581 * We don't want people slaying init.
583 * Note: we do this even if it is EXIT_DEAD, wait_task_zombie()
584 * can change ->exit_state to EXIT_ZOMBIE. If this is the final
585 * state, do_notify_parent() was already called and ->exit_signal
588 p->exit_signal = SIGCHLD;
590 if (p->exit_state == EXIT_DEAD)
593 /* If it has exited notify the new parent about this child's death. */
595 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
596 if (do_notify_parent(p, p->exit_signal)) {
597 p->exit_state = EXIT_DEAD;
598 list_move_tail(&p->sibling, dead);
602 kill_orphaned_pgrp(p, father);
605 static void forget_original_parent(struct task_struct *father)
607 struct task_struct *p, *n, *reaper;
608 LIST_HEAD(dead_children);
610 write_lock_irq(&tasklist_lock);
612 * Note that exit_ptrace() and find_new_reaper() might
613 * drop tasklist_lock and reacquire it.
616 reaper = find_new_reaper(father);
618 list_for_each_entry_safe(p, n, &father->children, sibling) {
619 struct task_struct *t = p;
621 t->real_parent = reaper;
622 if (t->parent == father) {
624 t->parent = t->real_parent;
626 if (t->pdeath_signal)
627 group_send_sig_info(t->pdeath_signal,
629 } while_each_thread(p, t);
630 reparent_leader(father, p, &dead_children);
632 write_unlock_irq(&tasklist_lock);
634 BUG_ON(!list_empty(&father->children));
636 list_for_each_entry_safe(p, n, &dead_children, sibling) {
637 list_del_init(&p->sibling);
643 * Send signals to all our closest relatives so that they know
644 * to properly mourn us..
646 static void exit_notify(struct task_struct *tsk, int group_dead)
651 * This does two things:
653 * A. Make init inherit all the child processes
654 * B. Check to see if any process groups have become orphaned
655 * as a result of our exiting, and if they have any stopped
656 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
658 forget_original_parent(tsk);
660 write_lock_irq(&tasklist_lock);
662 kill_orphaned_pgrp(tsk->group_leader, NULL);
664 if (unlikely(tsk->ptrace)) {
665 int sig = thread_group_leader(tsk) &&
666 thread_group_empty(tsk) &&
667 !ptrace_reparented(tsk) ?
668 tsk->exit_signal : SIGCHLD;
669 autoreap = do_notify_parent(tsk, sig);
670 } else if (thread_group_leader(tsk)) {
671 autoreap = thread_group_empty(tsk) &&
672 do_notify_parent(tsk, tsk->exit_signal);
677 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
679 /* mt-exec, de_thread() is waiting for group leader */
680 if (unlikely(tsk->signal->notify_count < 0))
681 wake_up_process(tsk->signal->group_exit_task);
682 write_unlock_irq(&tasklist_lock);
684 /* If the process is dead, release it - nobody will wait for it */
689 #ifdef CONFIG_DEBUG_STACK_USAGE
690 static void check_stack_usage(void)
692 static DEFINE_SPINLOCK(low_water_lock);
693 static int lowest_to_date = THREAD_SIZE;
696 free = stack_not_used(current);
698 if (free >= lowest_to_date)
701 spin_lock(&low_water_lock);
702 if (free < lowest_to_date) {
703 printk(KERN_WARNING "%s (%d) used greatest stack depth: "
705 current->comm, task_pid_nr(current), free);
706 lowest_to_date = free;
708 spin_unlock(&low_water_lock);
711 static inline void check_stack_usage(void) {}
714 void do_exit(long code)
716 struct task_struct *tsk = current;
719 profile_task_exit(tsk);
721 WARN_ON(blk_needs_flush_plug(tsk));
723 if (unlikely(in_interrupt()))
724 panic("Aiee, killing interrupt handler!");
725 if (unlikely(!tsk->pid))
726 panic("Attempted to kill the idle task!");
729 * If do_exit is called because this processes oopsed, it's possible
730 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
731 * continuing. Amongst other possible reasons, this is to prevent
732 * mm_release()->clear_child_tid() from writing to a user-controlled
737 ptrace_event(PTRACE_EVENT_EXIT, code);
739 validate_creds_for_do_exit(tsk);
742 * We're taking recursive faults here in do_exit. Safest is to just
743 * leave this task alone and wait for reboot.
745 if (unlikely(tsk->flags & PF_EXITING)) {
747 "Fixing recursive fault but reboot is needed!\n");
749 * We can do this unlocked here. The futex code uses
750 * this flag just to verify whether the pi state
751 * cleanup has been done or not. In the worst case it
752 * loops once more. We pretend that the cleanup was
753 * done as there is no way to return. Either the
754 * OWNER_DIED bit is set by now or we push the blocked
755 * task into the wait for ever nirwana as well.
757 tsk->flags |= PF_EXITPIDONE;
758 set_current_state(TASK_UNINTERRUPTIBLE);
762 exit_signals(tsk); /* sets PF_EXITING */
764 * tsk->flags are checked in the futex code to protect against
765 * an exiting task cleaning up the robust pi futexes.
768 raw_spin_unlock_wait(&tsk->pi_lock);
770 if (unlikely(in_atomic()))
771 printk(KERN_INFO "note: %s[%d] exited with preempt_count %d\n",
772 current->comm, task_pid_nr(current),
775 acct_update_integrals(tsk);
776 /* sync mm's RSS info before statistics gathering */
778 sync_mm_rss(tsk->mm);
779 group_dead = atomic_dec_and_test(&tsk->signal->live);
781 hrtimer_cancel(&tsk->signal->real_timer);
782 exit_itimers(tsk->signal);
784 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
786 acct_collect(code, group_dead);
791 tsk->exit_code = code;
792 taskstats_exit(tsk, group_dead);
798 trace_sched_process_exit(tsk);
805 disassociate_ctty(1);
806 exit_task_namespaces(tsk);
812 * Flush inherited counters to the parent - before the parent
813 * gets woken up by child-exit notifications.
815 * because of cgroup mode, must be called before cgroup_exit()
817 perf_event_exit_task(tsk);
821 module_put(task_thread_info(tsk)->exec_domain->module);
823 proc_exit_connector(tsk);
825 * FIXME: do that only when needed, using sched_exit tracepoint
827 ptrace_put_breakpoints(tsk);
829 exit_notify(tsk, group_dead);
832 mpol_put(tsk->mempolicy);
833 tsk->mempolicy = NULL;
837 if (unlikely(current->pi_state_cache))
838 kfree(current->pi_state_cache);
841 * Make sure we are holding no locks:
843 debug_check_no_locks_held(tsk);
845 * We can do this unlocked here. The futex code uses this flag
846 * just to verify whether the pi state cleanup has been done
847 * or not. In the worst case it loops once more.
849 tsk->flags |= PF_EXITPIDONE;
852 exit_io_context(tsk);
854 if (tsk->splice_pipe)
855 free_pipe_info(tsk->splice_pipe);
857 if (tsk->task_frag.page)
858 put_page(tsk->task_frag.page);
860 validate_creds_for_do_exit(tsk);
864 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
868 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
869 * when the following two conditions become true.
870 * - There is race condition of mmap_sem (It is acquired by
872 * - SMI occurs before setting TASK_RUNINNG.
873 * (or hypervisor of virtual machine switches to other guest)
874 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
876 * To avoid it, we have to wait for releasing tsk->pi_lock which
877 * is held by try_to_wake_up()
880 raw_spin_unlock_wait(&tsk->pi_lock);
882 /* causes final put_task_struct in finish_task_switch(). */
883 tsk->state = TASK_DEAD;
884 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
887 /* Avoid "noreturn function does return". */
889 cpu_relax(); /* For when BUG is null */
892 EXPORT_SYMBOL_GPL(do_exit);
894 void complete_and_exit(struct completion *comp, long code)
902 EXPORT_SYMBOL(complete_and_exit);
904 SYSCALL_DEFINE1(exit, int, error_code)
906 do_exit((error_code&0xff)<<8);
910 * Take down every thread in the group. This is called by fatal signals
911 * as well as by sys_exit_group (below).
914 do_group_exit(int exit_code)
916 struct signal_struct *sig = current->signal;
918 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
920 if (signal_group_exit(sig))
921 exit_code = sig->group_exit_code;
922 else if (!thread_group_empty(current)) {
923 struct sighand_struct *const sighand = current->sighand;
924 spin_lock_irq(&sighand->siglock);
925 if (signal_group_exit(sig))
926 /* Another thread got here before we took the lock. */
927 exit_code = sig->group_exit_code;
929 sig->group_exit_code = exit_code;
930 sig->flags = SIGNAL_GROUP_EXIT;
931 zap_other_threads(current);
933 spin_unlock_irq(&sighand->siglock);
941 * this kills every thread in the thread group. Note that any externally
942 * wait4()-ing process will get the correct exit code - even if this
943 * thread is not the thread group leader.
945 SYSCALL_DEFINE1(exit_group, int, error_code)
947 do_group_exit((error_code & 0xff) << 8);
953 enum pid_type wo_type;
957 struct siginfo __user *wo_info;
959 struct rusage __user *wo_rusage;
961 wait_queue_t child_wait;
966 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
968 if (type != PIDTYPE_PID)
969 task = task->group_leader;
970 return task->pids[type].pid;
973 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
975 return wo->wo_type == PIDTYPE_MAX ||
976 task_pid_type(p, wo->wo_type) == wo->wo_pid;
979 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
981 if (!eligible_pid(wo, p))
983 /* Wait for all children (clone and not) if __WALL is set;
984 * otherwise, wait for clone children *only* if __WCLONE is
985 * set; otherwise, wait for non-clone children *only*. (Note:
986 * A "clone" child here is one that reports to its parent
987 * using a signal other than SIGCHLD.) */
988 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
989 && !(wo->wo_flags & __WALL))
995 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
996 pid_t pid, uid_t uid, int why, int status)
998 struct siginfo __user *infop;
999 int retval = wo->wo_rusage
1000 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1003 infop = wo->wo_info;
1006 retval = put_user(SIGCHLD, &infop->si_signo);
1008 retval = put_user(0, &infop->si_errno);
1010 retval = put_user((short)why, &infop->si_code);
1012 retval = put_user(pid, &infop->si_pid);
1014 retval = put_user(uid, &infop->si_uid);
1016 retval = put_user(status, &infop->si_status);
1024 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
1025 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1026 * the lock and this task is uninteresting. If we return nonzero, we have
1027 * released the lock and the system call should return.
1029 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1031 unsigned long state;
1032 int retval, status, traced;
1033 pid_t pid = task_pid_vnr(p);
1034 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1035 struct siginfo __user *infop;
1037 if (!likely(wo->wo_flags & WEXITED))
1040 if (unlikely(wo->wo_flags & WNOWAIT)) {
1041 int exit_code = p->exit_code;
1045 read_unlock(&tasklist_lock);
1046 if ((exit_code & 0x7f) == 0) {
1048 status = exit_code >> 8;
1050 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1051 status = exit_code & 0x7f;
1053 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1057 * Try to move the task's state to DEAD
1058 * only one thread is allowed to do this:
1060 state = xchg(&p->exit_state, EXIT_DEAD);
1061 if (state != EXIT_ZOMBIE) {
1062 BUG_ON(state != EXIT_DEAD);
1066 traced = ptrace_reparented(p);
1068 * It can be ptraced but not reparented, check
1069 * thread_group_leader() to filter out sub-threads.
1071 if (likely(!traced) && thread_group_leader(p)) {
1072 struct signal_struct *psig;
1073 struct signal_struct *sig;
1074 unsigned long maxrss;
1075 cputime_t tgutime, tgstime;
1078 * The resource counters for the group leader are in its
1079 * own task_struct. Those for dead threads in the group
1080 * are in its signal_struct, as are those for the child
1081 * processes it has previously reaped. All these
1082 * accumulate in the parent's signal_struct c* fields.
1084 * We don't bother to take a lock here to protect these
1085 * p->signal fields, because they are only touched by
1086 * __exit_signal, which runs with tasklist_lock
1087 * write-locked anyway, and so is excluded here. We do
1088 * need to protect the access to parent->signal fields,
1089 * as other threads in the parent group can be right
1090 * here reaping other children at the same time.
1092 * We use thread_group_cputime_adjusted() to get times for the thread
1093 * group, which consolidates times for all threads in the
1094 * group including the group leader.
1096 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1097 spin_lock_irq(&p->real_parent->sighand->siglock);
1098 psig = p->real_parent->signal;
1100 psig->cutime += tgutime + sig->cutime;
1101 psig->cstime += tgstime + sig->cstime;
1102 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1104 p->min_flt + sig->min_flt + sig->cmin_flt;
1106 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1108 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1110 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1112 task_io_get_inblock(p) +
1113 sig->inblock + sig->cinblock;
1115 task_io_get_oublock(p) +
1116 sig->oublock + sig->coublock;
1117 maxrss = max(sig->maxrss, sig->cmaxrss);
1118 if (psig->cmaxrss < maxrss)
1119 psig->cmaxrss = maxrss;
1120 task_io_accounting_add(&psig->ioac, &p->ioac);
1121 task_io_accounting_add(&psig->ioac, &sig->ioac);
1122 spin_unlock_irq(&p->real_parent->sighand->siglock);
1126 * Now we are sure this task is interesting, and no other
1127 * thread can reap it because we set its state to EXIT_DEAD.
1129 read_unlock(&tasklist_lock);
1131 retval = wo->wo_rusage
1132 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1133 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1134 ? p->signal->group_exit_code : p->exit_code;
1135 if (!retval && wo->wo_stat)
1136 retval = put_user(status, wo->wo_stat);
1138 infop = wo->wo_info;
1139 if (!retval && infop)
1140 retval = put_user(SIGCHLD, &infop->si_signo);
1141 if (!retval && infop)
1142 retval = put_user(0, &infop->si_errno);
1143 if (!retval && infop) {
1146 if ((status & 0x7f) == 0) {
1150 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1153 retval = put_user((short)why, &infop->si_code);
1155 retval = put_user(status, &infop->si_status);
1157 if (!retval && infop)
1158 retval = put_user(pid, &infop->si_pid);
1159 if (!retval && infop)
1160 retval = put_user(uid, &infop->si_uid);
1165 write_lock_irq(&tasklist_lock);
1166 /* We dropped tasklist, ptracer could die and untrace */
1169 * If this is not a sub-thread, notify the parent.
1170 * If parent wants a zombie, don't release it now.
1172 if (thread_group_leader(p) &&
1173 !do_notify_parent(p, p->exit_signal)) {
1174 p->exit_state = EXIT_ZOMBIE;
1177 write_unlock_irq(&tasklist_lock);
1185 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1188 if (task_is_stopped_or_traced(p) &&
1189 !(p->jobctl & JOBCTL_LISTENING))
1190 return &p->exit_code;
1192 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1193 return &p->signal->group_exit_code;
1199 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1201 * @ptrace: is the wait for ptrace
1202 * @p: task to wait for
1204 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1207 * read_lock(&tasklist_lock), which is released if return value is
1208 * non-zero. Also, grabs and releases @p->sighand->siglock.
1211 * 0 if wait condition didn't exist and search for other wait conditions
1212 * should continue. Non-zero return, -errno on failure and @p's pid on
1213 * success, implies that tasklist_lock is released and wait condition
1214 * search should terminate.
1216 static int wait_task_stopped(struct wait_opts *wo,
1217 int ptrace, struct task_struct *p)
1219 struct siginfo __user *infop;
1220 int retval, exit_code, *p_code, why;
1221 uid_t uid = 0; /* unneeded, required by compiler */
1225 * Traditionally we see ptrace'd stopped tasks regardless of options.
1227 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1230 if (!task_stopped_code(p, ptrace))
1234 spin_lock_irq(&p->sighand->siglock);
1236 p_code = task_stopped_code(p, ptrace);
1237 if (unlikely(!p_code))
1240 exit_code = *p_code;
1244 if (!unlikely(wo->wo_flags & WNOWAIT))
1247 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1249 spin_unlock_irq(&p->sighand->siglock);
1254 * Now we are pretty sure this task is interesting.
1255 * Make sure it doesn't get reaped out from under us while we
1256 * give up the lock and then examine it below. We don't want to
1257 * keep holding onto the tasklist_lock while we call getrusage and
1258 * possibly take page faults for user memory.
1261 pid = task_pid_vnr(p);
1262 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1263 read_unlock(&tasklist_lock);
1265 if (unlikely(wo->wo_flags & WNOWAIT))
1266 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1268 retval = wo->wo_rusage
1269 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1270 if (!retval && wo->wo_stat)
1271 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1273 infop = wo->wo_info;
1274 if (!retval && infop)
1275 retval = put_user(SIGCHLD, &infop->si_signo);
1276 if (!retval && infop)
1277 retval = put_user(0, &infop->si_errno);
1278 if (!retval && infop)
1279 retval = put_user((short)why, &infop->si_code);
1280 if (!retval && infop)
1281 retval = put_user(exit_code, &infop->si_status);
1282 if (!retval && infop)
1283 retval = put_user(pid, &infop->si_pid);
1284 if (!retval && infop)
1285 retval = put_user(uid, &infop->si_uid);
1295 * Handle do_wait work for one task in a live, non-stopped state.
1296 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1297 * the lock and this task is uninteresting. If we return nonzero, we have
1298 * released the lock and the system call should return.
1300 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1306 if (!unlikely(wo->wo_flags & WCONTINUED))
1309 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1312 spin_lock_irq(&p->sighand->siglock);
1313 /* Re-check with the lock held. */
1314 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1315 spin_unlock_irq(&p->sighand->siglock);
1318 if (!unlikely(wo->wo_flags & WNOWAIT))
1319 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1320 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1321 spin_unlock_irq(&p->sighand->siglock);
1323 pid = task_pid_vnr(p);
1325 read_unlock(&tasklist_lock);
1328 retval = wo->wo_rusage
1329 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1331 if (!retval && wo->wo_stat)
1332 retval = put_user(0xffff, wo->wo_stat);
1336 retval = wait_noreap_copyout(wo, p, pid, uid,
1337 CLD_CONTINUED, SIGCONT);
1338 BUG_ON(retval == 0);
1345 * Consider @p for a wait by @parent.
1347 * -ECHILD should be in ->notask_error before the first call.
1348 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1349 * Returns zero if the search for a child should continue;
1350 * then ->notask_error is 0 if @p is an eligible child,
1351 * or another error from security_task_wait(), or still -ECHILD.
1353 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1354 struct task_struct *p)
1356 int ret = eligible_child(wo, p);
1360 ret = security_task_wait(p);
1361 if (unlikely(ret < 0)) {
1363 * If we have not yet seen any eligible child,
1364 * then let this error code replace -ECHILD.
1365 * A permission error will give the user a clue
1366 * to look for security policy problems, rather
1367 * than for mysterious wait bugs.
1369 if (wo->notask_error)
1370 wo->notask_error = ret;
1374 /* dead body doesn't have much to contribute */
1375 if (unlikely(p->exit_state == EXIT_DEAD)) {
1377 * But do not ignore this task until the tracer does
1378 * wait_task_zombie()->do_notify_parent().
1380 if (likely(!ptrace) && unlikely(ptrace_reparented(p)))
1381 wo->notask_error = 0;
1386 if (p->exit_state == EXIT_ZOMBIE) {
1388 * A zombie ptracee is only visible to its ptracer.
1389 * Notification and reaping will be cascaded to the real
1390 * parent when the ptracer detaches.
1392 if (likely(!ptrace) && unlikely(p->ptrace)) {
1393 /* it will become visible, clear notask_error */
1394 wo->notask_error = 0;
1398 /* we don't reap group leaders with subthreads */
1399 if (!delay_group_leader(p))
1400 return wait_task_zombie(wo, p);
1403 * Allow access to stopped/continued state via zombie by
1404 * falling through. Clearing of notask_error is complex.
1408 * If WEXITED is set, notask_error should naturally be
1409 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1410 * so, if there are live subthreads, there are events to
1411 * wait for. If all subthreads are dead, it's still safe
1412 * to clear - this function will be called again in finite
1413 * amount time once all the subthreads are released and
1414 * will then return without clearing.
1418 * Stopped state is per-task and thus can't change once the
1419 * target task dies. Only continued and exited can happen.
1420 * Clear notask_error if WCONTINUED | WEXITED.
1422 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1423 wo->notask_error = 0;
1426 * If @p is ptraced by a task in its real parent's group,
1427 * hide group stop/continued state when looking at @p as
1428 * the real parent; otherwise, a single stop can be
1429 * reported twice as group and ptrace stops.
1431 * If a ptracer wants to distinguish the two events for its
1432 * own children, it should create a separate process which
1433 * takes the role of real parent.
1435 if (likely(!ptrace) && p->ptrace && !ptrace_reparented(p))
1439 * @p is alive and it's gonna stop, continue or exit, so
1440 * there always is something to wait for.
1442 wo->notask_error = 0;
1446 * Wait for stopped. Depending on @ptrace, different stopped state
1447 * is used and the two don't interact with each other.
1449 ret = wait_task_stopped(wo, ptrace, p);
1454 * Wait for continued. There's only one continued state and the
1455 * ptracer can consume it which can confuse the real parent. Don't
1456 * use WCONTINUED from ptracer. You don't need or want it.
1458 return wait_task_continued(wo, p);
1462 * Do the work of do_wait() for one thread in the group, @tsk.
1464 * -ECHILD should be in ->notask_error before the first call.
1465 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1466 * Returns zero if the search for a child should continue; then
1467 * ->notask_error is 0 if there were any eligible children,
1468 * or another error from security_task_wait(), or still -ECHILD.
1470 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1472 struct task_struct *p;
1474 list_for_each_entry(p, &tsk->children, sibling) {
1475 int ret = wait_consider_task(wo, 0, p);
1483 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1485 struct task_struct *p;
1487 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1488 int ret = wait_consider_task(wo, 1, p);
1496 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1497 int sync, void *key)
1499 struct wait_opts *wo = container_of(wait, struct wait_opts,
1501 struct task_struct *p = key;
1503 if (!eligible_pid(wo, p))
1506 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1509 return default_wake_function(wait, mode, sync, key);
1512 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1514 __wake_up_sync_key(&parent->signal->wait_chldexit,
1515 TASK_INTERRUPTIBLE, 1, p);
1518 static long do_wait(struct wait_opts *wo)
1520 struct task_struct *tsk;
1523 trace_sched_process_wait(wo->wo_pid);
1525 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1526 wo->child_wait.private = current;
1527 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1530 * If there is nothing that can match our critiera just get out.
1531 * We will clear ->notask_error to zero if we see any child that
1532 * might later match our criteria, even if we are not able to reap
1535 wo->notask_error = -ECHILD;
1536 if ((wo->wo_type < PIDTYPE_MAX) &&
1537 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1540 set_current_state(TASK_INTERRUPTIBLE);
1541 read_lock(&tasklist_lock);
1544 retval = do_wait_thread(wo, tsk);
1548 retval = ptrace_do_wait(wo, tsk);
1552 if (wo->wo_flags & __WNOTHREAD)
1554 } while_each_thread(current, tsk);
1555 read_unlock(&tasklist_lock);
1558 retval = wo->notask_error;
1559 if (!retval && !(wo->wo_flags & WNOHANG)) {
1560 retval = -ERESTARTSYS;
1561 if (!signal_pending(current)) {
1567 __set_current_state(TASK_RUNNING);
1568 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1572 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1573 infop, int, options, struct rusage __user *, ru)
1575 struct wait_opts wo;
1576 struct pid *pid = NULL;
1580 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1582 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1595 type = PIDTYPE_PGID;
1603 if (type < PIDTYPE_MAX)
1604 pid = find_get_pid(upid);
1608 wo.wo_flags = options;
1618 * For a WNOHANG return, clear out all the fields
1619 * we would set so the user can easily tell the
1623 ret = put_user(0, &infop->si_signo);
1625 ret = put_user(0, &infop->si_errno);
1627 ret = put_user(0, &infop->si_code);
1629 ret = put_user(0, &infop->si_pid);
1631 ret = put_user(0, &infop->si_uid);
1633 ret = put_user(0, &infop->si_status);
1640 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1641 int, options, struct rusage __user *, ru)
1643 struct wait_opts wo;
1644 struct pid *pid = NULL;
1648 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1649 __WNOTHREAD|__WCLONE|__WALL))
1654 else if (upid < 0) {
1655 type = PIDTYPE_PGID;
1656 pid = find_get_pid(-upid);
1657 } else if (upid == 0) {
1658 type = PIDTYPE_PGID;
1659 pid = get_task_pid(current, PIDTYPE_PGID);
1660 } else /* upid > 0 */ {
1662 pid = find_get_pid(upid);
1667 wo.wo_flags = options | WEXITED;
1669 wo.wo_stat = stat_addr;
1677 #ifdef __ARCH_WANT_SYS_WAITPID
1680 * sys_waitpid() remains for compatibility. waitpid() should be
1681 * implemented by calling sys_wait4() from libc.a.
1683 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1685 return sys_wait4(pid, stat_addr, options, NULL);