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);
77 list_del_rcu(&p->thread_node);
81 * This function expects the tasklist_lock write-locked.
83 static void __exit_signal(struct task_struct *tsk)
85 struct signal_struct *sig = tsk->signal;
86 bool group_dead = thread_group_leader(tsk);
87 struct sighand_struct *sighand;
88 struct tty_struct *uninitialized_var(tty);
89 cputime_t utime, stime;
91 sighand = rcu_dereference_check(tsk->sighand,
92 lockdep_tasklist_lock_is_held());
93 spin_lock(&sighand->siglock);
95 posix_cpu_timers_exit(tsk);
97 posix_cpu_timers_exit_group(tsk);
102 * This can only happen if the caller is de_thread().
103 * FIXME: this is the temporary hack, we should teach
104 * posix-cpu-timers to handle this case correctly.
106 if (unlikely(has_group_leader_pid(tsk)))
107 posix_cpu_timers_exit_group(tsk);
110 * If there is any task waiting for the group exit
113 if (sig->notify_count > 0 && !--sig->notify_count)
114 wake_up_process(sig->group_exit_task);
116 if (tsk == sig->curr_target)
117 sig->curr_target = next_thread(tsk);
121 * Accumulate here the counters for all threads but the group leader
122 * as they die, so they can be added into the process-wide totals
123 * when those are taken. The group leader stays around as a zombie as
124 * long as there are other threads. When it gets reaped, the exit.c
125 * code will add its counts into these totals. We won't ever get here
126 * for the group leader, since it will have been the last reference on
129 task_cputime(tsk, &utime, &stime);
130 write_seqlock(&sig->stats_lock);
133 sig->gtime += task_gtime(tsk);
134 sig->min_flt += tsk->min_flt;
135 sig->maj_flt += tsk->maj_flt;
136 sig->nvcsw += tsk->nvcsw;
137 sig->nivcsw += tsk->nivcsw;
138 sig->inblock += task_io_get_inblock(tsk);
139 sig->oublock += task_io_get_oublock(tsk);
140 task_io_accounting_add(&sig->ioac, &tsk->ioac);
141 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
143 __unhash_process(tsk, group_dead);
144 write_sequnlock(&sig->stats_lock);
147 * Do this under ->siglock, we can race with another thread
148 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
150 flush_sigqueue(&tsk->pending);
152 spin_unlock(&sighand->siglock);
154 __cleanup_sighand(sighand);
155 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
157 flush_sigqueue(&sig->shared_pending);
162 static void delayed_put_task_struct(struct rcu_head *rhp)
164 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
166 perf_event_delayed_put(tsk);
167 trace_sched_process_free(tsk);
168 put_task_struct(tsk);
172 void release_task(struct task_struct *p)
174 struct task_struct *leader;
177 /* don't need to get the RCU readlock here - the process is dead and
178 * can't be modifying its own credentials. But shut RCU-lockdep up */
180 atomic_dec(&__task_cred(p)->user->processes);
185 write_lock_irq(&tasklist_lock);
186 ptrace_release_task(p);
190 * If we are the last non-leader member of the thread
191 * group, and the leader is zombie, then notify the
192 * group leader's parent process. (if it wants notification.)
195 leader = p->group_leader;
196 if (leader != p && thread_group_empty(leader)
197 && leader->exit_state == EXIT_ZOMBIE) {
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 zap_leader = do_notify_parent(leader, leader->exit_signal);
205 leader->exit_state = EXIT_DEAD;
208 write_unlock_irq(&tasklist_lock);
210 call_rcu(&p->rcu, delayed_put_task_struct);
213 if (unlikely(zap_leader))
218 * This checks not only the pgrp, but falls back on the pid if no
219 * satisfactory pgrp is found. I dunno - gdb doesn't work correctly
222 * The caller must hold rcu lock or the tasklist lock.
224 struct pid *session_of_pgrp(struct pid *pgrp)
226 struct task_struct *p;
227 struct pid *sid = NULL;
229 p = pid_task(pgrp, PIDTYPE_PGID);
231 p = pid_task(pgrp, PIDTYPE_PID);
233 sid = task_session(p);
239 * Determine if a process group is "orphaned", according to the POSIX
240 * definition in 2.2.2.52. Orphaned process groups are not to be affected
241 * by terminal-generated stop signals. Newly orphaned process groups are
242 * to receive a SIGHUP and a SIGCONT.
244 * "I ask you, have you ever known what it is to be an orphan?"
246 static int will_become_orphaned_pgrp(struct pid *pgrp,
247 struct task_struct *ignored_task)
249 struct task_struct *p;
251 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
252 if ((p == ignored_task) ||
253 (p->exit_state && thread_group_empty(p)) ||
254 is_global_init(p->real_parent))
257 if (task_pgrp(p->real_parent) != pgrp &&
258 task_session(p->real_parent) == task_session(p))
260 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
265 int is_current_pgrp_orphaned(void)
269 read_lock(&tasklist_lock);
270 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
271 read_unlock(&tasklist_lock);
276 static bool has_stopped_jobs(struct pid *pgrp)
278 struct task_struct *p;
280 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
281 if (p->signal->flags & SIGNAL_STOP_STOPPED)
283 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
289 * Check to see if any process groups have become orphaned as
290 * a result of our exiting, and if they have any stopped jobs,
291 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
294 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
296 struct pid *pgrp = task_pgrp(tsk);
297 struct task_struct *ignored_task = tsk;
300 /* exit: our father is in a different pgrp than
301 * we are and we were the only connection outside.
303 parent = tsk->real_parent;
305 /* reparent: our child is in a different pgrp than
306 * we are, and it was the only connection outside.
310 if (task_pgrp(parent) != pgrp &&
311 task_session(parent) == task_session(tsk) &&
312 will_become_orphaned_pgrp(pgrp, ignored_task) &&
313 has_stopped_jobs(pgrp)) {
314 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
315 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
321 * A task is exiting. If it owned this mm, find a new owner for the mm.
323 void mm_update_next_owner(struct mm_struct *mm)
325 struct task_struct *c, *g, *p = current;
329 * If the exiting or execing task is not the owner, it's
330 * someone else's problem.
335 * The current owner is exiting/execing and there are no other
336 * candidates. Do not leave the mm pointing to a possibly
337 * freed task structure.
339 if (atomic_read(&mm->mm_users) <= 1) {
344 read_lock(&tasklist_lock);
346 * Search in the children
348 list_for_each_entry(c, &p->children, sibling) {
350 goto assign_new_owner;
354 * Search in the siblings
356 list_for_each_entry(c, &p->real_parent->children, sibling) {
358 goto assign_new_owner;
362 * Search through everything else, we should not get here often.
364 for_each_process(g) {
365 if (g->flags & PF_KTHREAD)
367 for_each_thread(g, c) {
369 goto assign_new_owner;
374 read_unlock(&tasklist_lock);
376 * We found no owner yet mm_users > 1: this implies that we are
377 * most likely racing with swapoff (try_to_unuse()) or /proc or
378 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
387 * The task_lock protects c->mm from changing.
388 * We always want mm->owner->mm == mm
392 * Delay read_unlock() till we have the task_lock()
393 * to ensure that c does not slip away underneath us
395 read_unlock(&tasklist_lock);
405 #endif /* CONFIG_MEMCG */
408 * Turn us into a lazy TLB process if we
411 static void exit_mm(struct task_struct *tsk)
413 struct mm_struct *mm = tsk->mm;
414 struct core_state *core_state;
421 * Serialize with any possible pending coredump.
422 * We must hold mmap_sem around checking core_state
423 * and clearing tsk->mm. The core-inducing thread
424 * will increment ->nr_threads for each thread in the
425 * group with ->mm != NULL.
427 down_read(&mm->mmap_sem);
428 core_state = mm->core_state;
430 struct core_thread self;
432 up_read(&mm->mmap_sem);
435 self.next = xchg(&core_state->dumper.next, &self);
437 * Implies mb(), the result of xchg() must be visible
438 * to core_state->dumper.
440 if (atomic_dec_and_test(&core_state->nr_threads))
441 complete(&core_state->startup);
444 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
445 if (!self.task) /* see coredump_finish() */
447 freezable_schedule();
449 __set_task_state(tsk, TASK_RUNNING);
450 down_read(&mm->mmap_sem);
452 atomic_inc(&mm->mm_count);
453 BUG_ON(mm != tsk->active_mm);
454 /* more a memory barrier than a real lock */
457 up_read(&mm->mmap_sem);
458 enter_lazy_tlb(mm, current);
460 mm_update_next_owner(mm);
462 clear_thread_flag(TIF_MEMDIE);
466 * When we die, we re-parent all our children, and try to:
467 * 1. give them to another thread in our thread group, if such a member exists
468 * 2. give it to the first ancestor process which prctl'd itself as a
469 * child_subreaper for its children (like a service manager)
470 * 3. give it to the init process (PID 1) in our pid namespace
472 static struct task_struct *find_new_reaper(struct task_struct *father)
473 __releases(&tasklist_lock)
474 __acquires(&tasklist_lock)
476 struct pid_namespace *pid_ns = task_active_pid_ns(father);
477 struct task_struct *thread;
480 while_each_thread(father, thread) {
481 if (thread->flags & PF_EXITING)
483 if (unlikely(pid_ns->child_reaper == father))
484 pid_ns->child_reaper = thread;
488 if (unlikely(pid_ns->child_reaper == father)) {
489 write_unlock_irq(&tasklist_lock);
490 if (unlikely(pid_ns == &init_pid_ns)) {
491 panic("Attempted to kill init! exitcode=0x%08x\n",
492 father->signal->group_exit_code ?:
496 zap_pid_ns_processes(pid_ns);
497 write_lock_irq(&tasklist_lock);
498 } else if (father->signal->has_child_subreaper) {
499 struct task_struct *reaper;
502 * Find the first ancestor marked as child_subreaper.
503 * Note that the code below checks same_thread_group(reaper,
504 * pid_ns->child_reaper). This is what we need to DTRT in a
505 * PID namespace. However we still need the check above, see
506 * http://marc.info/?l=linux-kernel&m=131385460420380
508 for (reaper = father->real_parent;
509 reaper != &init_task;
510 reaper = reaper->real_parent) {
511 if (same_thread_group(reaper, pid_ns->child_reaper))
513 if (!reaper->signal->is_child_subreaper)
517 if (!(thread->flags & PF_EXITING))
519 } while_each_thread(reaper, thread);
523 return pid_ns->child_reaper;
527 * Any that need to be release_task'd are put on the @dead list.
529 static void reparent_leader(struct task_struct *father, struct task_struct *p,
530 struct list_head *dead)
532 if (unlikely(p->exit_state == EXIT_DEAD))
535 /* We don't want people slaying init. */
536 p->exit_signal = SIGCHLD;
538 /* If it has exited notify the new parent about this child's death. */
540 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
541 if (do_notify_parent(p, p->exit_signal)) {
542 p->exit_state = EXIT_DEAD;
543 list_add(&p->ptrace_entry, dead);
547 kill_orphaned_pgrp(p, father);
550 static void forget_original_parent(struct task_struct *father)
552 struct task_struct *p, *t, *n, *reaper;
553 LIST_HEAD(dead_children);
555 write_lock_irq(&tasklist_lock);
556 if (unlikely(!list_empty(&father->ptraced)))
557 exit_ptrace(father, &dead_children);
559 /* Can drop and reacquire tasklist_lock */
560 reaper = find_new_reaper(father);
561 list_for_each_entry(p, &father->children, sibling) {
562 for_each_thread(p, t) {
563 t->real_parent = reaper;
564 BUG_ON((!t->ptrace) != (t->parent == father));
565 if (likely(!t->ptrace))
566 t->parent = t->real_parent;
567 if (t->pdeath_signal)
568 group_send_sig_info(t->pdeath_signal,
572 * If this is a threaded reparent there is no need to
573 * notify anyone anything has happened.
575 if (!same_thread_group(reaper, father))
576 reparent_leader(father, p, &dead_children);
578 list_splice_tail_init(&father->children, &reaper->children);
579 write_unlock_irq(&tasklist_lock);
581 list_for_each_entry_safe(p, n, &dead_children, ptrace_entry) {
582 list_del_init(&p->ptrace_entry);
588 * Send signals to all our closest relatives so that they know
589 * to properly mourn us..
591 static void exit_notify(struct task_struct *tsk, int group_dead)
596 * This does two things:
598 * A. Make init inherit all the child processes
599 * B. Check to see if any process groups have become orphaned
600 * as a result of our exiting, and if they have any stopped
601 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
603 forget_original_parent(tsk);
605 write_lock_irq(&tasklist_lock);
607 kill_orphaned_pgrp(tsk->group_leader, NULL);
609 if (unlikely(tsk->ptrace)) {
610 int sig = thread_group_leader(tsk) &&
611 thread_group_empty(tsk) &&
612 !ptrace_reparented(tsk) ?
613 tsk->exit_signal : SIGCHLD;
614 autoreap = do_notify_parent(tsk, sig);
615 } else if (thread_group_leader(tsk)) {
616 autoreap = thread_group_empty(tsk) &&
617 do_notify_parent(tsk, tsk->exit_signal);
622 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
624 /* mt-exec, de_thread() is waiting for group leader */
625 if (unlikely(tsk->signal->notify_count < 0))
626 wake_up_process(tsk->signal->group_exit_task);
627 write_unlock_irq(&tasklist_lock);
629 /* If the process is dead, release it - nobody will wait for it */
634 #ifdef CONFIG_DEBUG_STACK_USAGE
635 static void check_stack_usage(void)
637 static DEFINE_SPINLOCK(low_water_lock);
638 static int lowest_to_date = THREAD_SIZE;
641 free = stack_not_used(current);
643 if (free >= lowest_to_date)
646 spin_lock(&low_water_lock);
647 if (free < lowest_to_date) {
648 pr_warn("%s (%d) used greatest stack depth: %lu bytes left\n",
649 current->comm, task_pid_nr(current), free);
650 lowest_to_date = free;
652 spin_unlock(&low_water_lock);
655 static inline void check_stack_usage(void) {}
658 void do_exit(long code)
660 struct task_struct *tsk = current;
662 TASKS_RCU(int tasks_rcu_i);
664 profile_task_exit(tsk);
666 WARN_ON(blk_needs_flush_plug(tsk));
668 if (unlikely(in_interrupt()))
669 panic("Aiee, killing interrupt handler!");
670 if (unlikely(!tsk->pid))
671 panic("Attempted to kill the idle task!");
674 * If do_exit is called because this processes oopsed, it's possible
675 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
676 * continuing. Amongst other possible reasons, this is to prevent
677 * mm_release()->clear_child_tid() from writing to a user-controlled
682 ptrace_event(PTRACE_EVENT_EXIT, code);
684 validate_creds_for_do_exit(tsk);
687 * We're taking recursive faults here in do_exit. Safest is to just
688 * leave this task alone and wait for reboot.
690 if (unlikely(tsk->flags & PF_EXITING)) {
691 pr_alert("Fixing recursive fault but reboot is needed!\n");
693 * We can do this unlocked here. The futex code uses
694 * this flag just to verify whether the pi state
695 * cleanup has been done or not. In the worst case it
696 * loops once more. We pretend that the cleanup was
697 * done as there is no way to return. Either the
698 * OWNER_DIED bit is set by now or we push the blocked
699 * task into the wait for ever nirwana as well.
701 tsk->flags |= PF_EXITPIDONE;
702 set_current_state(TASK_UNINTERRUPTIBLE);
706 exit_signals(tsk); /* sets PF_EXITING */
708 * tsk->flags are checked in the futex code to protect against
709 * an exiting task cleaning up the robust pi futexes.
712 raw_spin_unlock_wait(&tsk->pi_lock);
714 if (unlikely(in_atomic()))
715 pr_info("note: %s[%d] exited with preempt_count %d\n",
716 current->comm, task_pid_nr(current),
719 acct_update_integrals(tsk);
720 /* sync mm's RSS info before statistics gathering */
722 sync_mm_rss(tsk->mm);
723 group_dead = atomic_dec_and_test(&tsk->signal->live);
725 hrtimer_cancel(&tsk->signal->real_timer);
726 exit_itimers(tsk->signal);
728 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
730 acct_collect(code, group_dead);
735 tsk->exit_code = code;
736 taskstats_exit(tsk, group_dead);
742 trace_sched_process_exit(tsk);
749 disassociate_ctty(1);
750 exit_task_namespaces(tsk);
755 * Flush inherited counters to the parent - before the parent
756 * gets woken up by child-exit notifications.
758 * because of cgroup mode, must be called before cgroup_exit()
760 perf_event_exit_task(tsk);
764 module_put(task_thread_info(tsk)->exec_domain->module);
767 * FIXME: do that only when needed, using sched_exit tracepoint
769 flush_ptrace_hw_breakpoint(tsk);
771 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
772 exit_notify(tsk, group_dead);
773 proc_exit_connector(tsk);
776 mpol_put(tsk->mempolicy);
777 tsk->mempolicy = NULL;
781 if (unlikely(current->pi_state_cache))
782 kfree(current->pi_state_cache);
785 * Make sure we are holding no locks:
787 debug_check_no_locks_held();
789 * We can do this unlocked here. The futex code uses this flag
790 * just to verify whether the pi state cleanup has been done
791 * or not. In the worst case it loops once more.
793 tsk->flags |= PF_EXITPIDONE;
796 exit_io_context(tsk);
798 if (tsk->splice_pipe)
799 free_pipe_info(tsk->splice_pipe);
801 if (tsk->task_frag.page)
802 put_page(tsk->task_frag.page);
804 validate_creds_for_do_exit(tsk);
809 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
811 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
814 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
815 * when the following two conditions become true.
816 * - There is race condition of mmap_sem (It is acquired by
818 * - SMI occurs before setting TASK_RUNINNG.
819 * (or hypervisor of virtual machine switches to other guest)
820 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
822 * To avoid it, we have to wait for releasing tsk->pi_lock which
823 * is held by try_to_wake_up()
826 raw_spin_unlock_wait(&tsk->pi_lock);
828 /* causes final put_task_struct in finish_task_switch(). */
829 tsk->state = TASK_DEAD;
830 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
833 /* Avoid "noreturn function does return". */
835 cpu_relax(); /* For when BUG is null */
837 EXPORT_SYMBOL_GPL(do_exit);
839 void complete_and_exit(struct completion *comp, long code)
846 EXPORT_SYMBOL(complete_and_exit);
848 SYSCALL_DEFINE1(exit, int, error_code)
850 do_exit((error_code&0xff)<<8);
854 * Take down every thread in the group. This is called by fatal signals
855 * as well as by sys_exit_group (below).
858 do_group_exit(int exit_code)
860 struct signal_struct *sig = current->signal;
862 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
864 if (signal_group_exit(sig))
865 exit_code = sig->group_exit_code;
866 else if (!thread_group_empty(current)) {
867 struct sighand_struct *const sighand = current->sighand;
869 spin_lock_irq(&sighand->siglock);
870 if (signal_group_exit(sig))
871 /* Another thread got here before we took the lock. */
872 exit_code = sig->group_exit_code;
874 sig->group_exit_code = exit_code;
875 sig->flags = SIGNAL_GROUP_EXIT;
876 zap_other_threads(current);
878 spin_unlock_irq(&sighand->siglock);
886 * this kills every thread in the thread group. Note that any externally
887 * wait4()-ing process will get the correct exit code - even if this
888 * thread is not the thread group leader.
890 SYSCALL_DEFINE1(exit_group, int, error_code)
892 do_group_exit((error_code & 0xff) << 8);
898 enum pid_type wo_type;
902 struct siginfo __user *wo_info;
904 struct rusage __user *wo_rusage;
906 wait_queue_t child_wait;
911 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
913 if (type != PIDTYPE_PID)
914 task = task->group_leader;
915 return task->pids[type].pid;
918 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
920 return wo->wo_type == PIDTYPE_MAX ||
921 task_pid_type(p, wo->wo_type) == wo->wo_pid;
924 static int eligible_child(struct wait_opts *wo, struct task_struct *p)
926 if (!eligible_pid(wo, p))
928 /* Wait for all children (clone and not) if __WALL is set;
929 * otherwise, wait for clone children *only* if __WCLONE is
930 * set; otherwise, wait for non-clone children *only*. (Note:
931 * A "clone" child here is one that reports to its parent
932 * using a signal other than SIGCHLD.) */
933 if (((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
934 && !(wo->wo_flags & __WALL))
940 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
941 pid_t pid, uid_t uid, int why, int status)
943 struct siginfo __user *infop;
944 int retval = wo->wo_rusage
945 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
951 retval = put_user(SIGCHLD, &infop->si_signo);
953 retval = put_user(0, &infop->si_errno);
955 retval = put_user((short)why, &infop->si_code);
957 retval = put_user(pid, &infop->si_pid);
959 retval = put_user(uid, &infop->si_uid);
961 retval = put_user(status, &infop->si_status);
969 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
970 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
971 * the lock and this task is uninteresting. If we return nonzero, we have
972 * released the lock and the system call should return.
974 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
976 int state, retval, status;
977 pid_t pid = task_pid_vnr(p);
978 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
979 struct siginfo __user *infop;
981 if (!likely(wo->wo_flags & WEXITED))
984 if (unlikely(wo->wo_flags & WNOWAIT)) {
985 int exit_code = p->exit_code;
989 read_unlock(&tasklist_lock);
990 sched_annotate_sleep();
992 if ((exit_code & 0x7f) == 0) {
994 status = exit_code >> 8;
996 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
997 status = exit_code & 0x7f;
999 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1002 * Move the task's state to DEAD/TRACE, only one thread can do this.
1004 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1005 EXIT_TRACE : EXIT_DEAD;
1006 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1010 * Check thread_group_leader() to exclude the traced sub-threads.
1012 if (state == EXIT_DEAD && thread_group_leader(p)) {
1013 struct signal_struct *psig;
1014 struct signal_struct *sig;
1015 unsigned long maxrss;
1016 cputime_t tgutime, tgstime;
1019 * The resource counters for the group leader are in its
1020 * own task_struct. Those for dead threads in the group
1021 * are in its signal_struct, as are those for the child
1022 * processes it has previously reaped. All these
1023 * accumulate in the parent's signal_struct c* fields.
1025 * We don't bother to take a lock here to protect these
1026 * p->signal fields, because they are only touched by
1027 * __exit_signal, which runs with tasklist_lock
1028 * write-locked anyway, and so is excluded here. We do
1029 * need to protect the access to parent->signal fields,
1030 * as other threads in the parent group can be right
1031 * here reaping other children at the same time.
1033 * We use thread_group_cputime_adjusted() to get times for
1034 * the thread group, which consolidates times for all threads
1035 * in the group including the group leader.
1037 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1038 spin_lock_irq(&p->real_parent->sighand->siglock);
1039 psig = p->real_parent->signal;
1041 write_seqlock(&psig->stats_lock);
1042 psig->cutime += tgutime + sig->cutime;
1043 psig->cstime += tgstime + sig->cstime;
1044 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1046 p->min_flt + sig->min_flt + sig->cmin_flt;
1048 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1050 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1052 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1054 task_io_get_inblock(p) +
1055 sig->inblock + sig->cinblock;
1057 task_io_get_oublock(p) +
1058 sig->oublock + sig->coublock;
1059 maxrss = max(sig->maxrss, sig->cmaxrss);
1060 if (psig->cmaxrss < maxrss)
1061 psig->cmaxrss = maxrss;
1062 task_io_accounting_add(&psig->ioac, &p->ioac);
1063 task_io_accounting_add(&psig->ioac, &sig->ioac);
1064 write_sequnlock(&psig->stats_lock);
1065 spin_unlock_irq(&p->real_parent->sighand->siglock);
1069 * Now we are sure this task is interesting, and no other
1070 * thread can reap it because we its state == DEAD/TRACE.
1072 read_unlock(&tasklist_lock);
1073 sched_annotate_sleep();
1075 retval = wo->wo_rusage
1076 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1077 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1078 ? p->signal->group_exit_code : p->exit_code;
1079 if (!retval && wo->wo_stat)
1080 retval = put_user(status, wo->wo_stat);
1082 infop = wo->wo_info;
1083 if (!retval && infop)
1084 retval = put_user(SIGCHLD, &infop->si_signo);
1085 if (!retval && infop)
1086 retval = put_user(0, &infop->si_errno);
1087 if (!retval && infop) {
1090 if ((status & 0x7f) == 0) {
1094 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1097 retval = put_user((short)why, &infop->si_code);
1099 retval = put_user(status, &infop->si_status);
1101 if (!retval && infop)
1102 retval = put_user(pid, &infop->si_pid);
1103 if (!retval && infop)
1104 retval = put_user(uid, &infop->si_uid);
1108 if (state == EXIT_TRACE) {
1109 write_lock_irq(&tasklist_lock);
1110 /* We dropped tasklist, ptracer could die and untrace */
1113 /* If parent wants a zombie, don't release it now */
1114 state = EXIT_ZOMBIE;
1115 if (do_notify_parent(p, p->exit_signal))
1117 p->exit_state = state;
1118 write_unlock_irq(&tasklist_lock);
1120 if (state == EXIT_DEAD)
1126 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1129 if (task_is_stopped_or_traced(p) &&
1130 !(p->jobctl & JOBCTL_LISTENING))
1131 return &p->exit_code;
1133 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1134 return &p->signal->group_exit_code;
1140 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1142 * @ptrace: is the wait for ptrace
1143 * @p: task to wait for
1145 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1148 * read_lock(&tasklist_lock), which is released if return value is
1149 * non-zero. Also, grabs and releases @p->sighand->siglock.
1152 * 0 if wait condition didn't exist and search for other wait conditions
1153 * should continue. Non-zero return, -errno on failure and @p's pid on
1154 * success, implies that tasklist_lock is released and wait condition
1155 * search should terminate.
1157 static int wait_task_stopped(struct wait_opts *wo,
1158 int ptrace, struct task_struct *p)
1160 struct siginfo __user *infop;
1161 int retval, exit_code, *p_code, why;
1162 uid_t uid = 0; /* unneeded, required by compiler */
1166 * Traditionally we see ptrace'd stopped tasks regardless of options.
1168 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1171 if (!task_stopped_code(p, ptrace))
1175 spin_lock_irq(&p->sighand->siglock);
1177 p_code = task_stopped_code(p, ptrace);
1178 if (unlikely(!p_code))
1181 exit_code = *p_code;
1185 if (!unlikely(wo->wo_flags & WNOWAIT))
1188 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1190 spin_unlock_irq(&p->sighand->siglock);
1195 * Now we are pretty sure this task is interesting.
1196 * Make sure it doesn't get reaped out from under us while we
1197 * give up the lock and then examine it below. We don't want to
1198 * keep holding onto the tasklist_lock while we call getrusage and
1199 * possibly take page faults for user memory.
1202 pid = task_pid_vnr(p);
1203 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1204 read_unlock(&tasklist_lock);
1205 sched_annotate_sleep();
1207 if (unlikely(wo->wo_flags & WNOWAIT))
1208 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1210 retval = wo->wo_rusage
1211 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1212 if (!retval && wo->wo_stat)
1213 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1215 infop = wo->wo_info;
1216 if (!retval && infop)
1217 retval = put_user(SIGCHLD, &infop->si_signo);
1218 if (!retval && infop)
1219 retval = put_user(0, &infop->si_errno);
1220 if (!retval && infop)
1221 retval = put_user((short)why, &infop->si_code);
1222 if (!retval && infop)
1223 retval = put_user(exit_code, &infop->si_status);
1224 if (!retval && infop)
1225 retval = put_user(pid, &infop->si_pid);
1226 if (!retval && infop)
1227 retval = put_user(uid, &infop->si_uid);
1237 * Handle do_wait work for one task in a live, non-stopped state.
1238 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1239 * the lock and this task is uninteresting. If we return nonzero, we have
1240 * released the lock and the system call should return.
1242 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1248 if (!unlikely(wo->wo_flags & WCONTINUED))
1251 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1254 spin_lock_irq(&p->sighand->siglock);
1255 /* Re-check with the lock held. */
1256 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1257 spin_unlock_irq(&p->sighand->siglock);
1260 if (!unlikely(wo->wo_flags & WNOWAIT))
1261 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1262 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1263 spin_unlock_irq(&p->sighand->siglock);
1265 pid = task_pid_vnr(p);
1267 read_unlock(&tasklist_lock);
1268 sched_annotate_sleep();
1271 retval = wo->wo_rusage
1272 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1274 if (!retval && wo->wo_stat)
1275 retval = put_user(0xffff, wo->wo_stat);
1279 retval = wait_noreap_copyout(wo, p, pid, uid,
1280 CLD_CONTINUED, SIGCONT);
1281 BUG_ON(retval == 0);
1288 * Consider @p for a wait by @parent.
1290 * -ECHILD should be in ->notask_error before the first call.
1291 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1292 * Returns zero if the search for a child should continue;
1293 * then ->notask_error is 0 if @p is an eligible child,
1294 * or another error from security_task_wait(), or still -ECHILD.
1296 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1297 struct task_struct *p)
1301 if (unlikely(p->exit_state == EXIT_DEAD))
1304 ret = eligible_child(wo, p);
1308 ret = security_task_wait(p);
1309 if (unlikely(ret < 0)) {
1311 * If we have not yet seen any eligible child,
1312 * then let this error code replace -ECHILD.
1313 * A permission error will give the user a clue
1314 * to look for security policy problems, rather
1315 * than for mysterious wait bugs.
1317 if (wo->notask_error)
1318 wo->notask_error = ret;
1322 if (unlikely(p->exit_state == EXIT_TRACE)) {
1324 * ptrace == 0 means we are the natural parent. In this case
1325 * we should clear notask_error, debugger will notify us.
1327 if (likely(!ptrace))
1328 wo->notask_error = 0;
1332 if (likely(!ptrace) && unlikely(p->ptrace)) {
1334 * If it is traced by its real parent's group, just pretend
1335 * the caller is ptrace_do_wait() and reap this child if it
1338 * This also hides group stop state from real parent; otherwise
1339 * a single stop can be reported twice as group and ptrace stop.
1340 * If a ptracer wants to distinguish these two events for its
1341 * own children it should create a separate process which takes
1342 * the role of real parent.
1344 if (!ptrace_reparented(p))
1349 if (p->exit_state == EXIT_ZOMBIE) {
1350 /* we don't reap group leaders with subthreads */
1351 if (!delay_group_leader(p)) {
1353 * A zombie ptracee is only visible to its ptracer.
1354 * Notification and reaping will be cascaded to the
1355 * real parent when the ptracer detaches.
1357 if (unlikely(ptrace) || likely(!p->ptrace))
1358 return wait_task_zombie(wo, p);
1362 * Allow access to stopped/continued state via zombie by
1363 * falling through. Clearing of notask_error is complex.
1367 * If WEXITED is set, notask_error should naturally be
1368 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1369 * so, if there are live subthreads, there are events to
1370 * wait for. If all subthreads are dead, it's still safe
1371 * to clear - this function will be called again in finite
1372 * amount time once all the subthreads are released and
1373 * will then return without clearing.
1377 * Stopped state is per-task and thus can't change once the
1378 * target task dies. Only continued and exited can happen.
1379 * Clear notask_error if WCONTINUED | WEXITED.
1381 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1382 wo->notask_error = 0;
1385 * @p is alive and it's gonna stop, continue or exit, so
1386 * there always is something to wait for.
1388 wo->notask_error = 0;
1392 * Wait for stopped. Depending on @ptrace, different stopped state
1393 * is used and the two don't interact with each other.
1395 ret = wait_task_stopped(wo, ptrace, p);
1400 * Wait for continued. There's only one continued state and the
1401 * ptracer can consume it which can confuse the real parent. Don't
1402 * use WCONTINUED from ptracer. You don't need or want it.
1404 return wait_task_continued(wo, p);
1408 * Do the work of do_wait() for one thread in the group, @tsk.
1410 * -ECHILD should be in ->notask_error before the first call.
1411 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1412 * Returns zero if the search for a child should continue; then
1413 * ->notask_error is 0 if there were any eligible children,
1414 * or another error from security_task_wait(), or still -ECHILD.
1416 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1418 struct task_struct *p;
1420 list_for_each_entry(p, &tsk->children, sibling) {
1421 int ret = wait_consider_task(wo, 0, p);
1430 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1432 struct task_struct *p;
1434 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1435 int ret = wait_consider_task(wo, 1, p);
1444 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1445 int sync, void *key)
1447 struct wait_opts *wo = container_of(wait, struct wait_opts,
1449 struct task_struct *p = key;
1451 if (!eligible_pid(wo, p))
1454 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1457 return default_wake_function(wait, mode, sync, key);
1460 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1462 __wake_up_sync_key(&parent->signal->wait_chldexit,
1463 TASK_INTERRUPTIBLE, 1, p);
1466 static long do_wait(struct wait_opts *wo)
1468 struct task_struct *tsk;
1471 trace_sched_process_wait(wo->wo_pid);
1473 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1474 wo->child_wait.private = current;
1475 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1478 * If there is nothing that can match our critiera just get out.
1479 * We will clear ->notask_error to zero if we see any child that
1480 * might later match our criteria, even if we are not able to reap
1483 wo->notask_error = -ECHILD;
1484 if ((wo->wo_type < PIDTYPE_MAX) &&
1485 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1488 set_current_state(TASK_INTERRUPTIBLE);
1489 read_lock(&tasklist_lock);
1492 retval = do_wait_thread(wo, tsk);
1496 retval = ptrace_do_wait(wo, tsk);
1500 if (wo->wo_flags & __WNOTHREAD)
1502 } while_each_thread(current, tsk);
1503 read_unlock(&tasklist_lock);
1506 retval = wo->notask_error;
1507 if (!retval && !(wo->wo_flags & WNOHANG)) {
1508 retval = -ERESTARTSYS;
1509 if (!signal_pending(current)) {
1515 __set_current_state(TASK_RUNNING);
1516 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1520 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1521 infop, int, options, struct rusage __user *, ru)
1523 struct wait_opts wo;
1524 struct pid *pid = NULL;
1528 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1530 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1543 type = PIDTYPE_PGID;
1551 if (type < PIDTYPE_MAX)
1552 pid = find_get_pid(upid);
1556 wo.wo_flags = options;
1566 * For a WNOHANG return, clear out all the fields
1567 * we would set so the user can easily tell the
1571 ret = put_user(0, &infop->si_signo);
1573 ret = put_user(0, &infop->si_errno);
1575 ret = put_user(0, &infop->si_code);
1577 ret = put_user(0, &infop->si_pid);
1579 ret = put_user(0, &infop->si_uid);
1581 ret = put_user(0, &infop->si_status);
1588 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1589 int, options, struct rusage __user *, ru)
1591 struct wait_opts wo;
1592 struct pid *pid = NULL;
1596 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1597 __WNOTHREAD|__WCLONE|__WALL))
1602 else if (upid < 0) {
1603 type = PIDTYPE_PGID;
1604 pid = find_get_pid(-upid);
1605 } else if (upid == 0) {
1606 type = PIDTYPE_PGID;
1607 pid = get_task_pid(current, PIDTYPE_PGID);
1608 } else /* upid > 0 */ {
1610 pid = find_get_pid(upid);
1615 wo.wo_flags = options | WEXITED;
1617 wo.wo_stat = stat_addr;
1625 #ifdef __ARCH_WANT_SYS_WAITPID
1628 * sys_waitpid() remains for compatibility. waitpid() should be
1629 * implemented by calling sys_wait4() from libc.a.
1631 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1633 return sys_wait4(pid, stat_addr, options, NULL);