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>
56 #include <linux/kcov.h>
58 #include "sched/tune.h"
60 #include <asm/uaccess.h>
61 #include <asm/unistd.h>
62 #include <asm/pgtable.h>
63 #include <asm/mmu_context.h>
65 static void exit_mm(struct task_struct *tsk);
67 static void __unhash_process(struct task_struct *p, bool group_dead)
70 detach_pid(p, PIDTYPE_PID);
72 detach_pid(p, PIDTYPE_PGID);
73 detach_pid(p, PIDTYPE_SID);
75 list_del_rcu(&p->tasks);
76 list_del_init(&p->sibling);
77 __this_cpu_dec(process_counts);
79 list_del_rcu(&p->thread_group);
80 list_del_rcu(&p->thread_node);
84 * This function expects the tasklist_lock write-locked.
86 static void __exit_signal(struct task_struct *tsk)
88 struct signal_struct *sig = tsk->signal;
89 bool group_dead = thread_group_leader(tsk);
90 struct sighand_struct *sighand;
91 struct tty_struct *uninitialized_var(tty);
92 cputime_t utime, stime;
94 sighand = rcu_dereference_check(tsk->sighand,
95 lockdep_tasklist_lock_is_held());
96 spin_lock(&sighand->siglock);
98 posix_cpu_timers_exit(tsk);
100 posix_cpu_timers_exit_group(tsk);
105 * This can only happen if the caller is de_thread().
106 * FIXME: this is the temporary hack, we should teach
107 * posix-cpu-timers to handle this case correctly.
109 if (unlikely(has_group_leader_pid(tsk)))
110 posix_cpu_timers_exit_group(tsk);
113 * If there is any task waiting for the group exit
116 if (sig->notify_count > 0 && !--sig->notify_count)
117 wake_up_process(sig->group_exit_task);
119 if (tsk == sig->curr_target)
120 sig->curr_target = next_thread(tsk);
124 * Accumulate here the counters for all threads as they die. We could
125 * skip the group leader because it is the last user of signal_struct,
126 * but we want to avoid the race with thread_group_cputime() which can
127 * see the empty ->thread_head list.
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 * Determine if a process group is "orphaned", according to the POSIX
219 * definition in 2.2.2.52. Orphaned process groups are not to be affected
220 * by terminal-generated stop signals. Newly orphaned process groups are
221 * to receive a SIGHUP and a SIGCONT.
223 * "I ask you, have you ever known what it is to be an orphan?"
225 static int will_become_orphaned_pgrp(struct pid *pgrp,
226 struct task_struct *ignored_task)
228 struct task_struct *p;
230 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
231 if ((p == ignored_task) ||
232 (p->exit_state && thread_group_empty(p)) ||
233 is_global_init(p->real_parent))
236 if (task_pgrp(p->real_parent) != pgrp &&
237 task_session(p->real_parent) == task_session(p))
239 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
244 int is_current_pgrp_orphaned(void)
248 read_lock(&tasklist_lock);
249 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
250 read_unlock(&tasklist_lock);
255 static bool has_stopped_jobs(struct pid *pgrp)
257 struct task_struct *p;
259 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
260 if (p->signal->flags & SIGNAL_STOP_STOPPED)
262 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
268 * Check to see if any process groups have become orphaned as
269 * a result of our exiting, and if they have any stopped jobs,
270 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
273 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
275 struct pid *pgrp = task_pgrp(tsk);
276 struct task_struct *ignored_task = tsk;
279 /* exit: our father is in a different pgrp than
280 * we are and we were the only connection outside.
282 parent = tsk->real_parent;
284 /* reparent: our child is in a different pgrp than
285 * we are, and it was the only connection outside.
289 if (task_pgrp(parent) != pgrp &&
290 task_session(parent) == task_session(tsk) &&
291 will_become_orphaned_pgrp(pgrp, ignored_task) &&
292 has_stopped_jobs(pgrp)) {
293 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
294 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
300 * A task is exiting. If it owned this mm, find a new owner for the mm.
302 void mm_update_next_owner(struct mm_struct *mm)
304 struct task_struct *c, *g, *p = current;
308 * If the exiting or execing task is not the owner, it's
309 * someone else's problem.
314 * The current owner is exiting/execing and there are no other
315 * candidates. Do not leave the mm pointing to a possibly
316 * freed task structure.
318 if (atomic_read(&mm->mm_users) <= 1) {
323 read_lock(&tasklist_lock);
325 * Search in the children
327 list_for_each_entry(c, &p->children, sibling) {
329 goto assign_new_owner;
333 * Search in the siblings
335 list_for_each_entry(c, &p->real_parent->children, sibling) {
337 goto assign_new_owner;
341 * Search through everything else, we should not get here often.
343 for_each_process(g) {
344 if (g->flags & PF_KTHREAD)
346 for_each_thread(g, c) {
348 goto assign_new_owner;
353 read_unlock(&tasklist_lock);
355 * We found no owner yet mm_users > 1: this implies that we are
356 * most likely racing with swapoff (try_to_unuse()) or /proc or
357 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
366 * The task_lock protects c->mm from changing.
367 * We always want mm->owner->mm == mm
371 * Delay read_unlock() till we have the task_lock()
372 * to ensure that c does not slip away underneath us
374 read_unlock(&tasklist_lock);
384 #endif /* CONFIG_MEMCG */
387 * Turn us into a lazy TLB process if we
390 static void exit_mm(struct task_struct *tsk)
392 struct mm_struct *mm = tsk->mm;
393 struct core_state *core_state;
400 * Serialize with any possible pending coredump.
401 * We must hold mmap_sem around checking core_state
402 * and clearing tsk->mm. The core-inducing thread
403 * will increment ->nr_threads for each thread in the
404 * group with ->mm != NULL.
406 down_read(&mm->mmap_sem);
407 core_state = mm->core_state;
409 struct core_thread self;
411 up_read(&mm->mmap_sem);
414 self.next = xchg(&core_state->dumper.next, &self);
416 * Implies mb(), the result of xchg() must be visible
417 * to core_state->dumper.
419 if (atomic_dec_and_test(&core_state->nr_threads))
420 complete(&core_state->startup);
423 set_task_state(tsk, TASK_UNINTERRUPTIBLE);
424 if (!self.task) /* see coredump_finish() */
426 freezable_schedule();
428 __set_task_state(tsk, TASK_RUNNING);
429 down_read(&mm->mmap_sem);
431 atomic_inc(&mm->mm_count);
432 BUG_ON(mm != tsk->active_mm);
433 /* more a memory barrier than a real lock */
436 up_read(&mm->mmap_sem);
437 enter_lazy_tlb(mm, current);
439 mm_update_next_owner(mm);
441 if (test_thread_flag(TIF_MEMDIE))
445 static struct task_struct *find_alive_thread(struct task_struct *p)
447 struct task_struct *t;
449 for_each_thread(p, t) {
450 if (!(t->flags & PF_EXITING))
456 static struct task_struct *find_child_reaper(struct task_struct *father)
457 __releases(&tasklist_lock)
458 __acquires(&tasklist_lock)
460 struct pid_namespace *pid_ns = task_active_pid_ns(father);
461 struct task_struct *reaper = pid_ns->child_reaper;
463 if (likely(reaper != father))
466 reaper = find_alive_thread(father);
468 pid_ns->child_reaper = reaper;
472 write_unlock_irq(&tasklist_lock);
473 if (unlikely(pid_ns == &init_pid_ns)) {
474 panic("Attempted to kill init! exitcode=0x%08x\n",
475 father->signal->group_exit_code ?: father->exit_code);
477 zap_pid_ns_processes(pid_ns);
478 write_lock_irq(&tasklist_lock);
484 * When we die, we re-parent all our children, and try to:
485 * 1. give them to another thread in our thread group, if such a member exists
486 * 2. give it to the first ancestor process which prctl'd itself as a
487 * child_subreaper for its children (like a service manager)
488 * 3. give it to the init process (PID 1) in our pid namespace
490 static struct task_struct *find_new_reaper(struct task_struct *father,
491 struct task_struct *child_reaper)
493 struct task_struct *thread, *reaper;
495 thread = find_alive_thread(father);
499 if (father->signal->has_child_subreaper) {
501 * Find the first ->is_child_subreaper ancestor in our pid_ns.
502 * We start from father to ensure we can not look into another
503 * namespace, this is safe because all its threads are dead.
505 for (reaper = father;
506 !same_thread_group(reaper, child_reaper);
507 reaper = reaper->real_parent) {
508 /* call_usermodehelper() descendants need this check */
509 if (reaper == &init_task)
511 if (!reaper->signal->is_child_subreaper)
513 thread = find_alive_thread(reaper);
523 * Any that need to be release_task'd are put on the @dead list.
525 static void reparent_leader(struct task_struct *father, struct task_struct *p,
526 struct list_head *dead)
528 if (unlikely(p->exit_state == EXIT_DEAD))
531 /* We don't want people slaying init. */
532 p->exit_signal = SIGCHLD;
534 /* If it has exited notify the new parent about this child's death. */
536 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
537 if (do_notify_parent(p, p->exit_signal)) {
538 p->exit_state = EXIT_DEAD;
539 list_add(&p->ptrace_entry, dead);
543 kill_orphaned_pgrp(p, father);
547 * This does two things:
549 * A. Make init inherit all the child processes
550 * B. Check to see if any process groups have become orphaned
551 * as a result of our exiting, and if they have any stopped
552 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
554 static void forget_original_parent(struct task_struct *father,
555 struct list_head *dead)
557 struct task_struct *p, *t, *reaper;
559 if (unlikely(!list_empty(&father->ptraced)))
560 exit_ptrace(father, dead);
562 /* Can drop and reacquire tasklist_lock */
563 reaper = find_child_reaper(father);
564 if (list_empty(&father->children))
567 reaper = find_new_reaper(father, reaper);
568 list_for_each_entry(p, &father->children, sibling) {
569 for_each_thread(p, t) {
570 t->real_parent = reaper;
571 BUG_ON((!t->ptrace) != (t->parent == father));
572 if (likely(!t->ptrace))
573 t->parent = t->real_parent;
574 if (t->pdeath_signal)
575 group_send_sig_info(t->pdeath_signal,
579 * If this is a threaded reparent there is no need to
580 * notify anyone anything has happened.
582 if (!same_thread_group(reaper, father))
583 reparent_leader(father, p, dead);
585 list_splice_tail_init(&father->children, &reaper->children);
589 * Send signals to all our closest relatives so that they know
590 * to properly mourn us..
592 static void exit_notify(struct task_struct *tsk, int group_dead)
595 struct task_struct *p, *n;
598 write_lock_irq(&tasklist_lock);
599 forget_original_parent(tsk, &dead);
602 kill_orphaned_pgrp(tsk->group_leader, NULL);
604 if (unlikely(tsk->ptrace)) {
605 int sig = thread_group_leader(tsk) &&
606 thread_group_empty(tsk) &&
607 !ptrace_reparented(tsk) ?
608 tsk->exit_signal : SIGCHLD;
609 autoreap = do_notify_parent(tsk, sig);
610 } else if (thread_group_leader(tsk)) {
611 autoreap = thread_group_empty(tsk) &&
612 do_notify_parent(tsk, tsk->exit_signal);
617 tsk->exit_state = autoreap ? EXIT_DEAD : EXIT_ZOMBIE;
618 if (tsk->exit_state == EXIT_DEAD)
619 list_add(&tsk->ptrace_entry, &dead);
621 /* mt-exec, de_thread() is waiting for group leader */
622 if (unlikely(tsk->signal->notify_count < 0))
623 wake_up_process(tsk->signal->group_exit_task);
624 write_unlock_irq(&tasklist_lock);
626 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
627 list_del_init(&p->ptrace_entry);
632 #ifdef CONFIG_DEBUG_STACK_USAGE
633 static void check_stack_usage(void)
635 static DEFINE_SPINLOCK(low_water_lock);
636 static int lowest_to_date = THREAD_SIZE;
639 free = stack_not_used(current);
641 if (free >= lowest_to_date)
644 spin_lock(&low_water_lock);
645 if (free < lowest_to_date) {
646 pr_warn("%s (%d) used greatest stack depth: %lu bytes left\n",
647 current->comm, task_pid_nr(current), free);
648 lowest_to_date = free;
650 spin_unlock(&low_water_lock);
653 static inline void check_stack_usage(void) {}
656 void do_exit(long code)
658 struct task_struct *tsk = current;
660 TASKS_RCU(int tasks_rcu_i);
662 profile_task_exit(tsk);
665 WARN_ON(blk_needs_flush_plug(tsk));
667 if (unlikely(in_interrupt()))
668 panic("Aiee, killing interrupt handler!");
669 if (unlikely(!tsk->pid))
670 panic("Attempted to kill the idle task!");
673 * If do_exit is called because this processes oopsed, it's possible
674 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
675 * continuing. Amongst other possible reasons, this is to prevent
676 * mm_release()->clear_child_tid() from writing to a user-controlled
681 ptrace_event(PTRACE_EVENT_EXIT, code);
683 validate_creds_for_do_exit(tsk);
686 * We're taking recursive faults here in do_exit. Safest is to just
687 * leave this task alone and wait for reboot.
689 if (unlikely(tsk->flags & PF_EXITING)) {
690 pr_alert("Fixing recursive fault but reboot is needed!\n");
692 * We can do this unlocked here. The futex code uses
693 * this flag just to verify whether the pi state
694 * cleanup has been done or not. In the worst case it
695 * loops once more. We pretend that the cleanup was
696 * done as there is no way to return. Either the
697 * OWNER_DIED bit is set by now or we push the blocked
698 * task into the wait for ever nirwana as well.
700 tsk->flags |= PF_EXITPIDONE;
701 set_current_state(TASK_UNINTERRUPTIBLE);
705 exit_signals(tsk); /* sets PF_EXITING */
707 schedtune_exit_task(tsk);
710 * tsk->flags are checked in the futex code to protect against
711 * an exiting task cleaning up the robust pi futexes.
714 raw_spin_unlock_wait(&tsk->pi_lock);
716 if (unlikely(in_atomic())) {
717 pr_info("note: %s[%d] exited with preempt_count %d\n",
718 current->comm, task_pid_nr(current),
720 preempt_count_set(PREEMPT_ENABLED);
723 /* sync mm's RSS info before statistics gathering */
725 sync_mm_rss(tsk->mm);
726 acct_update_integrals(tsk);
727 group_dead = atomic_dec_and_test(&tsk->signal->live);
729 hrtimer_cancel(&tsk->signal->real_timer);
730 exit_itimers(tsk->signal);
732 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
734 acct_collect(code, group_dead);
739 tsk->exit_code = code;
740 taskstats_exit(tsk, group_dead);
746 trace_sched_process_exit(tsk);
753 disassociate_ctty(1);
754 exit_task_namespaces(tsk);
759 * Flush inherited counters to the parent - before the parent
760 * gets woken up by child-exit notifications.
762 * because of cgroup mode, must be called before cgroup_exit()
764 perf_event_exit_task(tsk);
769 * FIXME: do that only when needed, using sched_exit tracepoint
771 flush_ptrace_hw_breakpoint(tsk);
773 TASKS_RCU(preempt_disable());
774 TASKS_RCU(tasks_rcu_i = __srcu_read_lock(&tasks_rcu_exit_srcu));
775 TASKS_RCU(preempt_enable());
776 exit_notify(tsk, group_dead);
777 proc_exit_connector(tsk);
780 mpol_put(tsk->mempolicy);
781 tsk->mempolicy = NULL;
785 if (unlikely(current->pi_state_cache))
786 kfree(current->pi_state_cache);
789 * Make sure we are holding no locks:
791 debug_check_no_locks_held();
793 * We can do this unlocked here. The futex code uses this flag
794 * just to verify whether the pi state cleanup has been done
795 * or not. In the worst case it loops once more.
797 tsk->flags |= PF_EXITPIDONE;
800 exit_io_context(tsk);
802 if (tsk->splice_pipe)
803 free_pipe_info(tsk->splice_pipe);
805 if (tsk->task_frag.page)
806 put_page(tsk->task_frag.page);
808 validate_creds_for_do_exit(tsk);
813 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
815 TASKS_RCU(__srcu_read_unlock(&tasks_rcu_exit_srcu, tasks_rcu_i));
818 * The setting of TASK_RUNNING by try_to_wake_up() may be delayed
819 * when the following two conditions become true.
820 * - There is race condition of mmap_sem (It is acquired by
822 * - SMI occurs before setting TASK_RUNINNG.
823 * (or hypervisor of virtual machine switches to other guest)
824 * As a result, we may become TASK_RUNNING after becoming TASK_DEAD
826 * To avoid it, we have to wait for releasing tsk->pi_lock which
827 * is held by try_to_wake_up()
830 raw_spin_unlock_wait(&tsk->pi_lock);
832 /* causes final put_task_struct in finish_task_switch(). */
833 tsk->state = TASK_DEAD;
834 tsk->flags |= PF_NOFREEZE; /* tell freezer to ignore us */
837 /* Avoid "noreturn function does return". */
839 cpu_relax(); /* For when BUG is null */
841 EXPORT_SYMBOL_GPL(do_exit);
843 void complete_and_exit(struct completion *comp, long code)
850 EXPORT_SYMBOL(complete_and_exit);
852 SYSCALL_DEFINE1(exit, int, error_code)
854 do_exit((error_code&0xff)<<8);
858 * Take down every thread in the group. This is called by fatal signals
859 * as well as by sys_exit_group (below).
862 do_group_exit(int exit_code)
864 struct signal_struct *sig = current->signal;
866 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
868 if (signal_group_exit(sig))
869 exit_code = sig->group_exit_code;
870 else if (!thread_group_empty(current)) {
871 struct sighand_struct *const sighand = current->sighand;
873 spin_lock_irq(&sighand->siglock);
874 if (signal_group_exit(sig))
875 /* Another thread got here before we took the lock. */
876 exit_code = sig->group_exit_code;
878 sig->group_exit_code = exit_code;
879 sig->flags = SIGNAL_GROUP_EXIT;
880 zap_other_threads(current);
882 spin_unlock_irq(&sighand->siglock);
890 * this kills every thread in the thread group. Note that any externally
891 * wait4()-ing process will get the correct exit code - even if this
892 * thread is not the thread group leader.
894 SYSCALL_DEFINE1(exit_group, int, error_code)
896 do_group_exit((error_code & 0xff) << 8);
902 enum pid_type wo_type;
906 struct siginfo __user *wo_info;
908 struct rusage __user *wo_rusage;
910 wait_queue_t child_wait;
915 struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
917 if (type != PIDTYPE_PID)
918 task = task->group_leader;
919 return task->pids[type].pid;
922 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
924 return wo->wo_type == PIDTYPE_MAX ||
925 task_pid_type(p, wo->wo_type) == wo->wo_pid;
929 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
931 if (!eligible_pid(wo, p))
935 * Wait for all children (clone and not) if __WALL is set or
936 * if it is traced by us.
938 if (ptrace || (wo->wo_flags & __WALL))
942 * Otherwise, wait for clone children *only* if __WCLONE is set;
943 * otherwise, wait for non-clone children *only*.
945 * Note: a "clone" child here is one that reports to its parent
946 * using a signal other than SIGCHLD, or a non-leader thread which
947 * we can only see if it is traced by us.
949 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
955 static int wait_noreap_copyout(struct wait_opts *wo, struct task_struct *p,
956 pid_t pid, uid_t uid, int why, int status)
958 struct siginfo __user *infop;
959 int retval = wo->wo_rusage
960 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
966 retval = put_user(SIGCHLD, &infop->si_signo);
968 retval = put_user(0, &infop->si_errno);
970 retval = put_user((short)why, &infop->si_code);
972 retval = put_user(pid, &infop->si_pid);
974 retval = put_user(uid, &infop->si_uid);
976 retval = put_user(status, &infop->si_status);
984 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
985 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
986 * the lock and this task is uninteresting. If we return nonzero, we have
987 * released the lock and the system call should return.
989 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
991 int state, retval, status;
992 pid_t pid = task_pid_vnr(p);
993 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
994 struct siginfo __user *infop;
996 if (!likely(wo->wo_flags & WEXITED))
999 if (unlikely(wo->wo_flags & WNOWAIT)) {
1000 int exit_code = p->exit_code;
1004 read_unlock(&tasklist_lock);
1005 sched_annotate_sleep();
1007 if ((exit_code & 0x7f) == 0) {
1009 status = exit_code >> 8;
1011 why = (exit_code & 0x80) ? CLD_DUMPED : CLD_KILLED;
1012 status = exit_code & 0x7f;
1014 return wait_noreap_copyout(wo, p, pid, uid, why, status);
1017 * Move the task's state to DEAD/TRACE, only one thread can do this.
1019 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1020 EXIT_TRACE : EXIT_DEAD;
1021 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1024 * We own this thread, nobody else can reap it.
1026 read_unlock(&tasklist_lock);
1027 sched_annotate_sleep();
1030 * Check thread_group_leader() to exclude the traced sub-threads.
1032 if (state == EXIT_DEAD && thread_group_leader(p)) {
1033 struct signal_struct *sig = p->signal;
1034 struct signal_struct *psig = current->signal;
1035 unsigned long maxrss;
1036 cputime_t tgutime, tgstime;
1039 * The resource counters for the group leader are in its
1040 * own task_struct. Those for dead threads in the group
1041 * are in its signal_struct, as are those for the child
1042 * processes it has previously reaped. All these
1043 * accumulate in the parent's signal_struct c* fields.
1045 * We don't bother to take a lock here to protect these
1046 * p->signal fields because the whole thread group is dead
1047 * and nobody can change them.
1049 * psig->stats_lock also protects us from our sub-theads
1050 * which can reap other children at the same time. Until
1051 * we change k_getrusage()-like users to rely on this lock
1052 * we have to take ->siglock as well.
1054 * We use thread_group_cputime_adjusted() to get times for
1055 * the thread group, which consolidates times for all threads
1056 * in the group including the group leader.
1058 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1059 spin_lock_irq(¤t->sighand->siglock);
1060 write_seqlock(&psig->stats_lock);
1061 psig->cutime += tgutime + sig->cutime;
1062 psig->cstime += tgstime + sig->cstime;
1063 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1065 p->min_flt + sig->min_flt + sig->cmin_flt;
1067 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1069 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1071 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1073 task_io_get_inblock(p) +
1074 sig->inblock + sig->cinblock;
1076 task_io_get_oublock(p) +
1077 sig->oublock + sig->coublock;
1078 maxrss = max(sig->maxrss, sig->cmaxrss);
1079 if (psig->cmaxrss < maxrss)
1080 psig->cmaxrss = maxrss;
1081 task_io_accounting_add(&psig->ioac, &p->ioac);
1082 task_io_accounting_add(&psig->ioac, &sig->ioac);
1083 write_sequnlock(&psig->stats_lock);
1084 spin_unlock_irq(¤t->sighand->siglock);
1087 retval = wo->wo_rusage
1088 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1089 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1090 ? p->signal->group_exit_code : p->exit_code;
1091 if (!retval && wo->wo_stat)
1092 retval = put_user(status, wo->wo_stat);
1094 infop = wo->wo_info;
1095 if (!retval && infop)
1096 retval = put_user(SIGCHLD, &infop->si_signo);
1097 if (!retval && infop)
1098 retval = put_user(0, &infop->si_errno);
1099 if (!retval && infop) {
1102 if ((status & 0x7f) == 0) {
1106 why = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1109 retval = put_user((short)why, &infop->si_code);
1111 retval = put_user(status, &infop->si_status);
1113 if (!retval && infop)
1114 retval = put_user(pid, &infop->si_pid);
1115 if (!retval && infop)
1116 retval = put_user(uid, &infop->si_uid);
1120 if (state == EXIT_TRACE) {
1121 write_lock_irq(&tasklist_lock);
1122 /* We dropped tasklist, ptracer could die and untrace */
1125 /* If parent wants a zombie, don't release it now */
1126 state = EXIT_ZOMBIE;
1127 if (do_notify_parent(p, p->exit_signal))
1129 p->exit_state = state;
1130 write_unlock_irq(&tasklist_lock);
1132 if (state == EXIT_DEAD)
1138 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1141 if (task_is_stopped_or_traced(p) &&
1142 !(p->jobctl & JOBCTL_LISTENING))
1143 return &p->exit_code;
1145 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1146 return &p->signal->group_exit_code;
1152 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1154 * @ptrace: is the wait for ptrace
1155 * @p: task to wait for
1157 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1160 * read_lock(&tasklist_lock), which is released if return value is
1161 * non-zero. Also, grabs and releases @p->sighand->siglock.
1164 * 0 if wait condition didn't exist and search for other wait conditions
1165 * should continue. Non-zero return, -errno on failure and @p's pid on
1166 * success, implies that tasklist_lock is released and wait condition
1167 * search should terminate.
1169 static int wait_task_stopped(struct wait_opts *wo,
1170 int ptrace, struct task_struct *p)
1172 struct siginfo __user *infop;
1173 int retval, exit_code, *p_code, why;
1174 uid_t uid = 0; /* unneeded, required by compiler */
1178 * Traditionally we see ptrace'd stopped tasks regardless of options.
1180 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1183 if (!task_stopped_code(p, ptrace))
1187 spin_lock_irq(&p->sighand->siglock);
1189 p_code = task_stopped_code(p, ptrace);
1190 if (unlikely(!p_code))
1193 exit_code = *p_code;
1197 if (!unlikely(wo->wo_flags & WNOWAIT))
1200 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1202 spin_unlock_irq(&p->sighand->siglock);
1207 * Now we are pretty sure this task is interesting.
1208 * Make sure it doesn't get reaped out from under us while we
1209 * give up the lock and then examine it below. We don't want to
1210 * keep holding onto the tasklist_lock while we call getrusage and
1211 * possibly take page faults for user memory.
1214 pid = task_pid_vnr(p);
1215 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1216 read_unlock(&tasklist_lock);
1217 sched_annotate_sleep();
1219 if (unlikely(wo->wo_flags & WNOWAIT))
1220 return wait_noreap_copyout(wo, p, pid, uid, why, exit_code);
1222 retval = wo->wo_rusage
1223 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1224 if (!retval && wo->wo_stat)
1225 retval = put_user((exit_code << 8) | 0x7f, wo->wo_stat);
1227 infop = wo->wo_info;
1228 if (!retval && infop)
1229 retval = put_user(SIGCHLD, &infop->si_signo);
1230 if (!retval && infop)
1231 retval = put_user(0, &infop->si_errno);
1232 if (!retval && infop)
1233 retval = put_user((short)why, &infop->si_code);
1234 if (!retval && infop)
1235 retval = put_user(exit_code, &infop->si_status);
1236 if (!retval && infop)
1237 retval = put_user(pid, &infop->si_pid);
1238 if (!retval && infop)
1239 retval = put_user(uid, &infop->si_uid);
1249 * Handle do_wait work for one task in a live, non-stopped state.
1250 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1251 * the lock and this task is uninteresting. If we return nonzero, we have
1252 * released the lock and the system call should return.
1254 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1260 if (!unlikely(wo->wo_flags & WCONTINUED))
1263 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1266 spin_lock_irq(&p->sighand->siglock);
1267 /* Re-check with the lock held. */
1268 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1269 spin_unlock_irq(&p->sighand->siglock);
1272 if (!unlikely(wo->wo_flags & WNOWAIT))
1273 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1274 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1275 spin_unlock_irq(&p->sighand->siglock);
1277 pid = task_pid_vnr(p);
1279 read_unlock(&tasklist_lock);
1280 sched_annotate_sleep();
1283 retval = wo->wo_rusage
1284 ? getrusage(p, RUSAGE_BOTH, wo->wo_rusage) : 0;
1286 if (!retval && wo->wo_stat)
1287 retval = put_user(0xffff, wo->wo_stat);
1291 retval = wait_noreap_copyout(wo, p, pid, uid,
1292 CLD_CONTINUED, SIGCONT);
1293 BUG_ON(retval == 0);
1300 * Consider @p for a wait by @parent.
1302 * -ECHILD should be in ->notask_error before the first call.
1303 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1304 * Returns zero if the search for a child should continue;
1305 * then ->notask_error is 0 if @p is an eligible child,
1306 * or another error from security_task_wait(), or still -ECHILD.
1308 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1309 struct task_struct *p)
1312 * We can race with wait_task_zombie() from another thread.
1313 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1314 * can't confuse the checks below.
1316 int exit_state = ACCESS_ONCE(p->exit_state);
1319 if (unlikely(exit_state == EXIT_DEAD))
1322 ret = eligible_child(wo, ptrace, p);
1326 ret = security_task_wait(p);
1327 if (unlikely(ret < 0)) {
1329 * If we have not yet seen any eligible child,
1330 * then let this error code replace -ECHILD.
1331 * A permission error will give the user a clue
1332 * to look for security policy problems, rather
1333 * than for mysterious wait bugs.
1335 if (wo->notask_error)
1336 wo->notask_error = ret;
1340 if (unlikely(exit_state == EXIT_TRACE)) {
1342 * ptrace == 0 means we are the natural parent. In this case
1343 * we should clear notask_error, debugger will notify us.
1345 if (likely(!ptrace))
1346 wo->notask_error = 0;
1350 if (likely(!ptrace) && unlikely(p->ptrace)) {
1352 * If it is traced by its real parent's group, just pretend
1353 * the caller is ptrace_do_wait() and reap this child if it
1356 * This also hides group stop state from real parent; otherwise
1357 * a single stop can be reported twice as group and ptrace stop.
1358 * If a ptracer wants to distinguish these two events for its
1359 * own children it should create a separate process which takes
1360 * the role of real parent.
1362 if (!ptrace_reparented(p))
1367 if (exit_state == EXIT_ZOMBIE) {
1368 /* we don't reap group leaders with subthreads */
1369 if (!delay_group_leader(p)) {
1371 * A zombie ptracee is only visible to its ptracer.
1372 * Notification and reaping will be cascaded to the
1373 * real parent when the ptracer detaches.
1375 if (unlikely(ptrace) || likely(!p->ptrace))
1376 return wait_task_zombie(wo, p);
1380 * Allow access to stopped/continued state via zombie by
1381 * falling through. Clearing of notask_error is complex.
1385 * If WEXITED is set, notask_error should naturally be
1386 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1387 * so, if there are live subthreads, there are events to
1388 * wait for. If all subthreads are dead, it's still safe
1389 * to clear - this function will be called again in finite
1390 * amount time once all the subthreads are released and
1391 * will then return without clearing.
1395 * Stopped state is per-task and thus can't change once the
1396 * target task dies. Only continued and exited can happen.
1397 * Clear notask_error if WCONTINUED | WEXITED.
1399 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1400 wo->notask_error = 0;
1403 * @p is alive and it's gonna stop, continue or exit, so
1404 * there always is something to wait for.
1406 wo->notask_error = 0;
1410 * Wait for stopped. Depending on @ptrace, different stopped state
1411 * is used and the two don't interact with each other.
1413 ret = wait_task_stopped(wo, ptrace, p);
1418 * Wait for continued. There's only one continued state and the
1419 * ptracer can consume it which can confuse the real parent. Don't
1420 * use WCONTINUED from ptracer. You don't need or want it.
1422 return wait_task_continued(wo, p);
1426 * Do the work of do_wait() for one thread in the group, @tsk.
1428 * -ECHILD should be in ->notask_error before the first call.
1429 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1430 * Returns zero if the search for a child should continue; then
1431 * ->notask_error is 0 if there were any eligible children,
1432 * or another error from security_task_wait(), or still -ECHILD.
1434 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1436 struct task_struct *p;
1438 list_for_each_entry(p, &tsk->children, sibling) {
1439 int ret = wait_consider_task(wo, 0, p);
1448 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1450 struct task_struct *p;
1452 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1453 int ret = wait_consider_task(wo, 1, p);
1462 static int child_wait_callback(wait_queue_t *wait, unsigned mode,
1463 int sync, void *key)
1465 struct wait_opts *wo = container_of(wait, struct wait_opts,
1467 struct task_struct *p = key;
1469 if (!eligible_pid(wo, p))
1472 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1475 return default_wake_function(wait, mode, sync, key);
1478 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1480 __wake_up_sync_key(&parent->signal->wait_chldexit,
1481 TASK_INTERRUPTIBLE, 1, p);
1484 static long do_wait(struct wait_opts *wo)
1486 struct task_struct *tsk;
1489 trace_sched_process_wait(wo->wo_pid);
1491 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1492 wo->child_wait.private = current;
1493 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1496 * If there is nothing that can match our criteria, just get out.
1497 * We will clear ->notask_error to zero if we see any child that
1498 * might later match our criteria, even if we are not able to reap
1501 wo->notask_error = -ECHILD;
1502 if ((wo->wo_type < PIDTYPE_MAX) &&
1503 (!wo->wo_pid || hlist_empty(&wo->wo_pid->tasks[wo->wo_type])))
1506 set_current_state(TASK_INTERRUPTIBLE);
1507 read_lock(&tasklist_lock);
1510 retval = do_wait_thread(wo, tsk);
1514 retval = ptrace_do_wait(wo, tsk);
1518 if (wo->wo_flags & __WNOTHREAD)
1520 } while_each_thread(current, tsk);
1521 read_unlock(&tasklist_lock);
1524 retval = wo->notask_error;
1525 if (!retval && !(wo->wo_flags & WNOHANG)) {
1526 retval = -ERESTARTSYS;
1527 if (!signal_pending(current)) {
1533 __set_current_state(TASK_RUNNING);
1534 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1538 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1539 infop, int, options, struct rusage __user *, ru)
1541 struct wait_opts wo;
1542 struct pid *pid = NULL;
1546 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED))
1548 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1561 type = PIDTYPE_PGID;
1569 if (type < PIDTYPE_MAX)
1570 pid = find_get_pid(upid);
1574 wo.wo_flags = options;
1584 * For a WNOHANG return, clear out all the fields
1585 * we would set so the user can easily tell the
1589 ret = put_user(0, &infop->si_signo);
1591 ret = put_user(0, &infop->si_errno);
1593 ret = put_user(0, &infop->si_code);
1595 ret = put_user(0, &infop->si_pid);
1597 ret = put_user(0, &infop->si_uid);
1599 ret = put_user(0, &infop->si_status);
1606 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1607 int, options, struct rusage __user *, ru)
1609 struct wait_opts wo;
1610 struct pid *pid = NULL;
1614 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1615 __WNOTHREAD|__WCLONE|__WALL))
1618 /* -INT_MIN is not defined */
1619 if (upid == INT_MIN)
1624 else if (upid < 0) {
1625 type = PIDTYPE_PGID;
1626 pid = find_get_pid(-upid);
1627 } else if (upid == 0) {
1628 type = PIDTYPE_PGID;
1629 pid = get_task_pid(current, PIDTYPE_PGID);
1630 } else /* upid > 0 */ {
1632 pid = find_get_pid(upid);
1637 wo.wo_flags = options | WEXITED;
1639 wo.wo_stat = stat_addr;
1647 #ifdef __ARCH_WANT_SYS_WAITPID
1650 * sys_waitpid() remains for compatibility. waitpid() should be
1651 * implemented by calling sys_wait4() from libc.a.
1653 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1655 return sys_wait4(pid, stat_addr, options, NULL);