1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 #include <linux/slab.h>
10 #include <linux/sched/autogroup.h>
11 #include <linux/sched/mm.h>
12 #include <linux/sched/stat.h>
13 #include <linux/sched/task.h>
14 #include <linux/sched/task_stack.h>
15 #include <linux/sched/cputime.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/capability.h>
19 #include <linux/completion.h>
20 #include <linux/personality.h>
21 #include <linux/tty.h>
22 #include <linux/iocontext.h>
23 #include <linux/key.h>
24 #include <linux/cpu.h>
25 #include <linux/acct.h>
26 #include <linux/tsacct_kern.h>
27 #include <linux/file.h>
28 #include <linux/fdtable.h>
29 #include <linux/freezer.h>
30 #include <linux/binfmts.h>
31 #include <linux/nsproxy.h>
32 #include <linux/pid_namespace.h>
33 #include <linux/ptrace.h>
34 #include <linux/profile.h>
35 #include <linux/mount.h>
36 #include <linux/proc_fs.h>
37 #include <linux/kthread.h>
38 #include <linux/mempolicy.h>
39 #include <linux/taskstats_kern.h>
40 #include <linux/delayacct.h>
41 #include <linux/cgroup.h>
42 #include <linux/syscalls.h>
43 #include <linux/signal.h>
44 #include <linux/posix-timers.h>
45 #include <linux/cn_proc.h>
46 #include <linux/mutex.h>
47 #include <linux/futex.h>
48 #include <linux/pipe_fs_i.h>
49 #include <linux/audit.h> /* for audit_free() */
50 #include <linux/resource.h>
51 #include <linux/blkdev.h>
52 #include <linux/task_io_accounting_ops.h>
53 #include <linux/tracehook.h>
54 #include <linux/fs_struct.h>
55 #include <linux/init_task.h>
56 #include <linux/perf_event.h>
57 #include <trace/events/sched.h>
58 #include <linux/hw_breakpoint.h>
59 #include <linux/oom.h>
60 #include <linux/writeback.h>
61 #include <linux/shm.h>
62 #include <linux/kcov.h>
63 #include <linux/random.h>
64 #include <linux/rcuwait.h>
65 #include <linux/compat.h>
66 #include <linux/io_uring.h>
68 #include <linux/uaccess.h>
69 #include <asm/unistd.h>
70 #include <asm/mmu_context.h>
72 static void __unhash_process(struct task_struct *p, bool group_dead)
75 detach_pid(p, PIDTYPE_PID);
77 detach_pid(p, PIDTYPE_TGID);
78 detach_pid(p, PIDTYPE_PGID);
79 detach_pid(p, PIDTYPE_SID);
81 list_del_rcu(&p->tasks);
82 list_del_init(&p->sibling);
83 __this_cpu_dec(process_counts);
85 list_del_rcu(&p->thread_group);
86 list_del_rcu(&p->thread_node);
90 * This function expects the tasklist_lock write-locked.
92 static void __exit_signal(struct task_struct *tsk)
94 struct signal_struct *sig = tsk->signal;
95 bool group_dead = thread_group_leader(tsk);
96 struct sighand_struct *sighand;
97 struct tty_struct *tty;
100 sighand = rcu_dereference_check(tsk->sighand,
101 lockdep_tasklist_lock_is_held());
102 spin_lock(&sighand->siglock);
104 #ifdef CONFIG_POSIX_TIMERS
105 posix_cpu_timers_exit(tsk);
107 posix_cpu_timers_exit_group(tsk);
115 * If there is any task waiting for the group exit
118 if (sig->notify_count > 0 && !--sig->notify_count)
119 wake_up_process(sig->group_exit_task);
121 if (tsk == sig->curr_target)
122 sig->curr_target = next_thread(tsk);
125 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
126 sizeof(unsigned long long));
129 * Accumulate here the counters for all threads as they die. We could
130 * skip the group leader because it is the last user of signal_struct,
131 * but we want to avoid the race with thread_group_cputime() which can
132 * see the empty ->thread_head list.
134 task_cputime(tsk, &utime, &stime);
135 write_seqlock(&sig->stats_lock);
138 sig->gtime += task_gtime(tsk);
139 sig->min_flt += tsk->min_flt;
140 sig->maj_flt += tsk->maj_flt;
141 sig->nvcsw += tsk->nvcsw;
142 sig->nivcsw += tsk->nivcsw;
143 sig->inblock += task_io_get_inblock(tsk);
144 sig->oublock += task_io_get_oublock(tsk);
145 task_io_accounting_add(&sig->ioac, &tsk->ioac);
146 sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
148 __unhash_process(tsk, group_dead);
149 write_sequnlock(&sig->stats_lock);
152 * Do this under ->siglock, we can race with another thread
153 * doing sigqueue_free() if we have SIGQUEUE_PREALLOC signals.
155 flush_sigqueue(&tsk->pending);
157 spin_unlock(&sighand->siglock);
159 __cleanup_sighand(sighand);
160 clear_tsk_thread_flag(tsk, TIF_SIGPENDING);
162 flush_sigqueue(&sig->shared_pending);
167 static void delayed_put_task_struct(struct rcu_head *rhp)
169 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
171 perf_event_delayed_put(tsk);
172 trace_sched_process_free(tsk);
173 put_task_struct(tsk);
176 void put_task_struct_rcu_user(struct task_struct *task)
178 if (refcount_dec_and_test(&task->rcu_users))
179 call_rcu(&task->rcu, delayed_put_task_struct);
182 void release_task(struct task_struct *p)
184 struct task_struct *leader;
185 struct pid *thread_pid;
188 /* don't need to get the RCU readlock here - the process is dead and
189 * can't be modifying its own credentials. But shut RCU-lockdep up */
191 dec_rlimit_ucounts(task_ucounts(p), UCOUNT_RLIMIT_NPROC, 1);
196 write_lock_irq(&tasklist_lock);
197 ptrace_release_task(p);
198 thread_pid = get_pid(p->thread_pid);
202 * If we are the last non-leader member of the thread
203 * group, and the leader is zombie, then notify the
204 * group leader's parent process. (if it wants notification.)
207 leader = p->group_leader;
208 if (leader != p && thread_group_empty(leader)
209 && leader->exit_state == EXIT_ZOMBIE) {
211 * If we were the last child thread and the leader has
212 * exited already, and the leader's parent ignores SIGCHLD,
213 * then we are the one who should release the leader.
215 zap_leader = do_notify_parent(leader, leader->exit_signal);
217 leader->exit_state = EXIT_DEAD;
220 write_unlock_irq(&tasklist_lock);
221 seccomp_filter_release(p);
222 proc_flush_pid(thread_pid);
225 put_task_struct_rcu_user(p);
228 if (unlikely(zap_leader))
232 int rcuwait_wake_up(struct rcuwait *w)
235 struct task_struct *task;
240 * Order condition vs @task, such that everything prior to the load
241 * of @task is visible. This is the condition as to why the user called
242 * rcuwait_wake() in the first place. Pairs with set_current_state()
243 * barrier (A) in rcuwait_wait_event().
246 * [S] tsk = current [S] cond = true
252 task = rcu_dereference(w->task);
254 ret = wake_up_process(task);
259 EXPORT_SYMBOL_GPL(rcuwait_wake_up);
262 * Determine if a process group is "orphaned", according to the POSIX
263 * definition in 2.2.2.52. Orphaned process groups are not to be affected
264 * by terminal-generated stop signals. Newly orphaned process groups are
265 * to receive a SIGHUP and a SIGCONT.
267 * "I ask you, have you ever known what it is to be an orphan?"
269 static int will_become_orphaned_pgrp(struct pid *pgrp,
270 struct task_struct *ignored_task)
272 struct task_struct *p;
274 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
275 if ((p == ignored_task) ||
276 (p->exit_state && thread_group_empty(p)) ||
277 is_global_init(p->real_parent))
280 if (task_pgrp(p->real_parent) != pgrp &&
281 task_session(p->real_parent) == task_session(p))
283 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
288 int is_current_pgrp_orphaned(void)
292 read_lock(&tasklist_lock);
293 retval = will_become_orphaned_pgrp(task_pgrp(current), NULL);
294 read_unlock(&tasklist_lock);
299 static bool has_stopped_jobs(struct pid *pgrp)
301 struct task_struct *p;
303 do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
304 if (p->signal->flags & SIGNAL_STOP_STOPPED)
306 } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
312 * Check to see if any process groups have become orphaned as
313 * a result of our exiting, and if they have any stopped jobs,
314 * send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
317 kill_orphaned_pgrp(struct task_struct *tsk, struct task_struct *parent)
319 struct pid *pgrp = task_pgrp(tsk);
320 struct task_struct *ignored_task = tsk;
323 /* exit: our father is in a different pgrp than
324 * we are and we were the only connection outside.
326 parent = tsk->real_parent;
328 /* reparent: our child is in a different pgrp than
329 * we are, and it was the only connection outside.
333 if (task_pgrp(parent) != pgrp &&
334 task_session(parent) == task_session(tsk) &&
335 will_become_orphaned_pgrp(pgrp, ignored_task) &&
336 has_stopped_jobs(pgrp)) {
337 __kill_pgrp_info(SIGHUP, SEND_SIG_PRIV, pgrp);
338 __kill_pgrp_info(SIGCONT, SEND_SIG_PRIV, pgrp);
344 * A task is exiting. If it owned this mm, find a new owner for the mm.
346 void mm_update_next_owner(struct mm_struct *mm)
348 struct task_struct *c, *g, *p = current;
352 * If the exiting or execing task is not the owner, it's
353 * someone else's problem.
358 * The current owner is exiting/execing and there are no other
359 * candidates. Do not leave the mm pointing to a possibly
360 * freed task structure.
362 if (atomic_read(&mm->mm_users) <= 1) {
363 WRITE_ONCE(mm->owner, NULL);
367 read_lock(&tasklist_lock);
369 * Search in the children
371 list_for_each_entry(c, &p->children, sibling) {
373 goto assign_new_owner;
377 * Search in the siblings
379 list_for_each_entry(c, &p->real_parent->children, sibling) {
381 goto assign_new_owner;
385 * Search through everything else, we should not get here often.
387 for_each_process(g) {
388 if (g->flags & PF_KTHREAD)
390 for_each_thread(g, c) {
392 goto assign_new_owner;
397 read_unlock(&tasklist_lock);
399 * We found no owner yet mm_users > 1: this implies that we are
400 * most likely racing with swapoff (try_to_unuse()) or /proc or
401 * ptrace or page migration (get_task_mm()). Mark owner as NULL.
403 WRITE_ONCE(mm->owner, NULL);
410 * The task_lock protects c->mm from changing.
411 * We always want mm->owner->mm == mm
415 * Delay read_unlock() till we have the task_lock()
416 * to ensure that c does not slip away underneath us
418 read_unlock(&tasklist_lock);
424 WRITE_ONCE(mm->owner, c);
428 #endif /* CONFIG_MEMCG */
431 * Turn us into a lazy TLB process if we
434 static void exit_mm(void)
436 struct mm_struct *mm = current->mm;
437 struct core_state *core_state;
439 exit_mm_release(current, mm);
444 * Serialize with any possible pending coredump.
445 * We must hold mmap_lock around checking core_state
446 * and clearing tsk->mm. The core-inducing thread
447 * will increment ->nr_threads for each thread in the
448 * group with ->mm != NULL.
451 core_state = mm->core_state;
453 struct core_thread self;
455 mmap_read_unlock(mm);
458 if (self.task->flags & PF_SIGNALED)
459 self.next = xchg(&core_state->dumper.next, &self);
463 * Implies mb(), the result of xchg() must be visible
464 * to core_state->dumper.
466 if (atomic_dec_and_test(&core_state->nr_threads))
467 complete(&core_state->startup);
470 set_current_state(TASK_UNINTERRUPTIBLE);
471 if (!self.task) /* see coredump_finish() */
473 freezable_schedule();
475 __set_current_state(TASK_RUNNING);
479 BUG_ON(mm != current->active_mm);
480 /* more a memory barrier than a real lock */
483 * When a thread stops operating on an address space, the loop
484 * in membarrier_private_expedited() may not observe that
485 * tsk->mm, and the loop in membarrier_global_expedited() may
486 * not observe a MEMBARRIER_STATE_GLOBAL_EXPEDITED
487 * rq->membarrier_state, so those would not issue an IPI.
488 * Membarrier requires a memory barrier after accessing
489 * user-space memory, before clearing tsk->mm or the
490 * rq->membarrier_state.
492 smp_mb__after_spinlock();
495 membarrier_update_current_mm(NULL);
496 enter_lazy_tlb(mm, current);
498 task_unlock(current);
499 mmap_read_unlock(mm);
500 mm_update_next_owner(mm);
502 if (test_thread_flag(TIF_MEMDIE))
506 static struct task_struct *find_alive_thread(struct task_struct *p)
508 struct task_struct *t;
510 for_each_thread(p, t) {
511 if (!(t->flags & PF_EXITING))
517 static struct task_struct *find_child_reaper(struct task_struct *father,
518 struct list_head *dead)
519 __releases(&tasklist_lock)
520 __acquires(&tasklist_lock)
522 struct pid_namespace *pid_ns = task_active_pid_ns(father);
523 struct task_struct *reaper = pid_ns->child_reaper;
524 struct task_struct *p, *n;
526 if (likely(reaper != father))
529 reaper = find_alive_thread(father);
531 pid_ns->child_reaper = reaper;
535 write_unlock_irq(&tasklist_lock);
537 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
538 list_del_init(&p->ptrace_entry);
542 zap_pid_ns_processes(pid_ns);
543 write_lock_irq(&tasklist_lock);
549 * When we die, we re-parent all our children, and try to:
550 * 1. give them to another thread in our thread group, if such a member exists
551 * 2. give it to the first ancestor process which prctl'd itself as a
552 * child_subreaper for its children (like a service manager)
553 * 3. give it to the init process (PID 1) in our pid namespace
555 static struct task_struct *find_new_reaper(struct task_struct *father,
556 struct task_struct *child_reaper)
558 struct task_struct *thread, *reaper;
560 thread = find_alive_thread(father);
564 if (father->signal->has_child_subreaper) {
565 unsigned int ns_level = task_pid(father)->level;
567 * Find the first ->is_child_subreaper ancestor in our pid_ns.
568 * We can't check reaper != child_reaper to ensure we do not
569 * cross the namespaces, the exiting parent could be injected
570 * by setns() + fork().
571 * We check pid->level, this is slightly more efficient than
572 * task_active_pid_ns(reaper) != task_active_pid_ns(father).
574 for (reaper = father->real_parent;
575 task_pid(reaper)->level == ns_level;
576 reaper = reaper->real_parent) {
577 if (reaper == &init_task)
579 if (!reaper->signal->is_child_subreaper)
581 thread = find_alive_thread(reaper);
591 * Any that need to be release_task'd are put on the @dead list.
593 static void reparent_leader(struct task_struct *father, struct task_struct *p,
594 struct list_head *dead)
596 if (unlikely(p->exit_state == EXIT_DEAD))
599 /* We don't want people slaying init. */
600 p->exit_signal = SIGCHLD;
602 /* If it has exited notify the new parent about this child's death. */
604 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
605 if (do_notify_parent(p, p->exit_signal)) {
606 p->exit_state = EXIT_DEAD;
607 list_add(&p->ptrace_entry, dead);
611 kill_orphaned_pgrp(p, father);
615 * This does two things:
617 * A. Make init inherit all the child processes
618 * B. Check to see if any process groups have become orphaned
619 * as a result of our exiting, and if they have any stopped
620 * jobs, send them a SIGHUP and then a SIGCONT. (POSIX 3.2.2.2)
622 static void forget_original_parent(struct task_struct *father,
623 struct list_head *dead)
625 struct task_struct *p, *t, *reaper;
627 if (unlikely(!list_empty(&father->ptraced)))
628 exit_ptrace(father, dead);
630 /* Can drop and reacquire tasklist_lock */
631 reaper = find_child_reaper(father, dead);
632 if (list_empty(&father->children))
635 reaper = find_new_reaper(father, reaper);
636 list_for_each_entry(p, &father->children, sibling) {
637 for_each_thread(p, t) {
638 RCU_INIT_POINTER(t->real_parent, reaper);
639 BUG_ON((!t->ptrace) != (rcu_access_pointer(t->parent) == father));
640 if (likely(!t->ptrace))
641 t->parent = t->real_parent;
642 if (t->pdeath_signal)
643 group_send_sig_info(t->pdeath_signal,
648 * If this is a threaded reparent there is no need to
649 * notify anyone anything has happened.
651 if (!same_thread_group(reaper, father))
652 reparent_leader(father, p, dead);
654 list_splice_tail_init(&father->children, &reaper->children);
658 * Send signals to all our closest relatives so that they know
659 * to properly mourn us..
661 static void exit_notify(struct task_struct *tsk, int group_dead)
664 struct task_struct *p, *n;
667 write_lock_irq(&tasklist_lock);
668 forget_original_parent(tsk, &dead);
671 kill_orphaned_pgrp(tsk->group_leader, NULL);
673 tsk->exit_state = EXIT_ZOMBIE;
674 if (unlikely(tsk->ptrace)) {
675 int sig = thread_group_leader(tsk) &&
676 thread_group_empty(tsk) &&
677 !ptrace_reparented(tsk) ?
678 tsk->exit_signal : SIGCHLD;
679 autoreap = do_notify_parent(tsk, sig);
680 } else if (thread_group_leader(tsk)) {
681 autoreap = thread_group_empty(tsk) &&
682 do_notify_parent(tsk, tsk->exit_signal);
688 tsk->exit_state = EXIT_DEAD;
689 list_add(&tsk->ptrace_entry, &dead);
692 /* mt-exec, de_thread() is waiting for group leader */
693 if (unlikely(tsk->signal->notify_count < 0))
694 wake_up_process(tsk->signal->group_exit_task);
695 write_unlock_irq(&tasklist_lock);
697 list_for_each_entry_safe(p, n, &dead, ptrace_entry) {
698 list_del_init(&p->ptrace_entry);
703 #ifdef CONFIG_DEBUG_STACK_USAGE
704 static void check_stack_usage(void)
706 static DEFINE_SPINLOCK(low_water_lock);
707 static int lowest_to_date = THREAD_SIZE;
710 free = stack_not_used(current);
712 if (free >= lowest_to_date)
715 spin_lock(&low_water_lock);
716 if (free < lowest_to_date) {
717 pr_info("%s (%d) used greatest stack depth: %lu bytes left\n",
718 current->comm, task_pid_nr(current), free);
719 lowest_to_date = free;
721 spin_unlock(&low_water_lock);
724 static inline void check_stack_usage(void) {}
727 void __noreturn do_exit(long code)
729 struct task_struct *tsk = current;
733 * We can get here from a kernel oops, sometimes with preemption off.
734 * Start by checking for critical errors.
735 * Then fix up important state like USER_DS and preemption.
736 * Then do everything else.
739 WARN_ON(blk_needs_flush_plug(tsk));
741 if (unlikely(in_interrupt()))
742 panic("Aiee, killing interrupt handler!");
743 if (unlikely(!tsk->pid))
744 panic("Attempted to kill the idle task!");
747 * If do_exit is called because this processes oopsed, it's possible
748 * that get_fs() was left as KERNEL_DS, so reset it to USER_DS before
749 * continuing. Amongst other possible reasons, this is to prevent
750 * mm_release()->clear_child_tid() from writing to a user-controlled
753 force_uaccess_begin();
755 if (unlikely(in_atomic())) {
756 pr_info("note: %s[%d] exited with preempt_count %d\n",
757 current->comm, task_pid_nr(current),
759 preempt_count_set(PREEMPT_ENABLED);
762 profile_task_exit(tsk);
765 ptrace_event(PTRACE_EVENT_EXIT, code);
767 validate_creds_for_do_exit(tsk);
770 * We're taking recursive faults here in do_exit. Safest is to just
771 * leave this task alone and wait for reboot.
773 if (unlikely(tsk->flags & PF_EXITING)) {
774 pr_alert("Fixing recursive fault but reboot is needed!\n");
775 futex_exit_recursive(tsk);
776 set_current_state(TASK_UNINTERRUPTIBLE);
780 io_uring_files_cancel();
781 exit_signals(tsk); /* sets PF_EXITING */
783 /* sync mm's RSS info before statistics gathering */
785 sync_mm_rss(tsk->mm);
786 acct_update_integrals(tsk);
787 group_dead = atomic_dec_and_test(&tsk->signal->live);
790 * If the last thread of global init has exited, panic
791 * immediately to get a useable coredump.
793 if (unlikely(is_global_init(tsk)))
794 panic("Attempted to kill init! exitcode=0x%08x\n",
795 tsk->signal->group_exit_code ?: (int)code);
797 #ifdef CONFIG_POSIX_TIMERS
798 hrtimer_cancel(&tsk->signal->real_timer);
799 exit_itimers(tsk->signal);
802 setmax_mm_hiwater_rss(&tsk->signal->maxrss, tsk->mm);
804 acct_collect(code, group_dead);
809 tsk->exit_code = code;
810 taskstats_exit(tsk, group_dead);
816 trace_sched_process_exit(tsk);
823 disassociate_ctty(1);
824 exit_task_namespaces(tsk);
829 * Flush inherited counters to the parent - before the parent
830 * gets woken up by child-exit notifications.
832 * because of cgroup mode, must be called before cgroup_exit()
834 perf_event_exit_task(tsk);
836 sched_autogroup_exit_task(tsk);
840 * FIXME: do that only when needed, using sched_exit tracepoint
842 flush_ptrace_hw_breakpoint(tsk);
844 exit_tasks_rcu_start();
845 exit_notify(tsk, group_dead);
846 proc_exit_connector(tsk);
847 mpol_put_task_policy(tsk);
849 if (unlikely(current->pi_state_cache))
850 kfree(current->pi_state_cache);
853 * Make sure we are holding no locks:
855 debug_check_no_locks_held();
858 exit_io_context(tsk);
860 if (tsk->splice_pipe)
861 free_pipe_info(tsk->splice_pipe);
863 if (tsk->task_frag.page)
864 put_page(tsk->task_frag.page);
866 validate_creds_for_do_exit(tsk);
871 __this_cpu_add(dirty_throttle_leaks, tsk->nr_dirtied);
873 exit_tasks_rcu_finish();
875 lockdep_free_task(tsk);
878 EXPORT_SYMBOL_GPL(do_exit);
880 void complete_and_exit(struct completion *comp, long code)
887 EXPORT_SYMBOL(complete_and_exit);
889 SYSCALL_DEFINE1(exit, int, error_code)
891 do_exit((error_code&0xff)<<8);
895 * Take down every thread in the group. This is called by fatal signals
896 * as well as by sys_exit_group (below).
899 do_group_exit(int exit_code)
901 struct signal_struct *sig = current->signal;
903 BUG_ON(exit_code & 0x80); /* core dumps don't get here */
905 if (signal_group_exit(sig))
906 exit_code = sig->group_exit_code;
907 else if (!thread_group_empty(current)) {
908 struct sighand_struct *const sighand = current->sighand;
910 spin_lock_irq(&sighand->siglock);
911 if (signal_group_exit(sig))
912 /* Another thread got here before we took the lock. */
913 exit_code = sig->group_exit_code;
915 sig->group_exit_code = exit_code;
916 sig->flags = SIGNAL_GROUP_EXIT;
917 zap_other_threads(current);
919 spin_unlock_irq(&sighand->siglock);
927 * this kills every thread in the thread group. Note that any externally
928 * wait4()-ing process will get the correct exit code - even if this
929 * thread is not the thread group leader.
931 SYSCALL_DEFINE1(exit_group, int, error_code)
933 do_group_exit((error_code & 0xff) << 8);
946 enum pid_type wo_type;
950 struct waitid_info *wo_info;
952 struct rusage *wo_rusage;
954 wait_queue_entry_t child_wait;
958 static int eligible_pid(struct wait_opts *wo, struct task_struct *p)
960 return wo->wo_type == PIDTYPE_MAX ||
961 task_pid_type(p, wo->wo_type) == wo->wo_pid;
965 eligible_child(struct wait_opts *wo, bool ptrace, struct task_struct *p)
967 if (!eligible_pid(wo, p))
971 * Wait for all children (clone and not) if __WALL is set or
972 * if it is traced by us.
974 if (ptrace || (wo->wo_flags & __WALL))
978 * Otherwise, wait for clone children *only* if __WCLONE is set;
979 * otherwise, wait for non-clone children *only*.
981 * Note: a "clone" child here is one that reports to its parent
982 * using a signal other than SIGCHLD, or a non-leader thread which
983 * we can only see if it is traced by us.
985 if ((p->exit_signal != SIGCHLD) ^ !!(wo->wo_flags & __WCLONE))
992 * Handle sys_wait4 work for one task in state EXIT_ZOMBIE. We hold
993 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
994 * the lock and this task is uninteresting. If we return nonzero, we have
995 * released the lock and the system call should return.
997 static int wait_task_zombie(struct wait_opts *wo, struct task_struct *p)
1000 pid_t pid = task_pid_vnr(p);
1001 uid_t uid = from_kuid_munged(current_user_ns(), task_uid(p));
1002 struct waitid_info *infop;
1004 if (!likely(wo->wo_flags & WEXITED))
1007 if (unlikely(wo->wo_flags & WNOWAIT)) {
1008 status = p->exit_code;
1010 read_unlock(&tasklist_lock);
1011 sched_annotate_sleep();
1013 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1018 * Move the task's state to DEAD/TRACE, only one thread can do this.
1020 state = (ptrace_reparented(p) && thread_group_leader(p)) ?
1021 EXIT_TRACE : EXIT_DEAD;
1022 if (cmpxchg(&p->exit_state, EXIT_ZOMBIE, state) != EXIT_ZOMBIE)
1025 * We own this thread, nobody else can reap it.
1027 read_unlock(&tasklist_lock);
1028 sched_annotate_sleep();
1031 * Check thread_group_leader() to exclude the traced sub-threads.
1033 if (state == EXIT_DEAD && thread_group_leader(p)) {
1034 struct signal_struct *sig = p->signal;
1035 struct signal_struct *psig = current->signal;
1036 unsigned long maxrss;
1037 u64 tgutime, tgstime;
1040 * The resource counters for the group leader are in its
1041 * own task_struct. Those for dead threads in the group
1042 * are in its signal_struct, as are those for the child
1043 * processes it has previously reaped. All these
1044 * accumulate in the parent's signal_struct c* fields.
1046 * We don't bother to take a lock here to protect these
1047 * p->signal fields because the whole thread group is dead
1048 * and nobody can change them.
1050 * psig->stats_lock also protects us from our sub-theads
1051 * which can reap other children at the same time. Until
1052 * we change k_getrusage()-like users to rely on this lock
1053 * we have to take ->siglock as well.
1055 * We use thread_group_cputime_adjusted() to get times for
1056 * the thread group, which consolidates times for all threads
1057 * in the group including the group leader.
1059 thread_group_cputime_adjusted(p, &tgutime, &tgstime);
1060 spin_lock_irq(¤t->sighand->siglock);
1061 write_seqlock(&psig->stats_lock);
1062 psig->cutime += tgutime + sig->cutime;
1063 psig->cstime += tgstime + sig->cstime;
1064 psig->cgtime += task_gtime(p) + sig->gtime + sig->cgtime;
1066 p->min_flt + sig->min_flt + sig->cmin_flt;
1068 p->maj_flt + sig->maj_flt + sig->cmaj_flt;
1070 p->nvcsw + sig->nvcsw + sig->cnvcsw;
1072 p->nivcsw + sig->nivcsw + sig->cnivcsw;
1074 task_io_get_inblock(p) +
1075 sig->inblock + sig->cinblock;
1077 task_io_get_oublock(p) +
1078 sig->oublock + sig->coublock;
1079 maxrss = max(sig->maxrss, sig->cmaxrss);
1080 if (psig->cmaxrss < maxrss)
1081 psig->cmaxrss = maxrss;
1082 task_io_accounting_add(&psig->ioac, &p->ioac);
1083 task_io_accounting_add(&psig->ioac, &sig->ioac);
1084 write_sequnlock(&psig->stats_lock);
1085 spin_unlock_irq(¤t->sighand->siglock);
1089 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1090 status = (p->signal->flags & SIGNAL_GROUP_EXIT)
1091 ? p->signal->group_exit_code : p->exit_code;
1092 wo->wo_stat = status;
1094 if (state == EXIT_TRACE) {
1095 write_lock_irq(&tasklist_lock);
1096 /* We dropped tasklist, ptracer could die and untrace */
1099 /* If parent wants a zombie, don't release it now */
1100 state = EXIT_ZOMBIE;
1101 if (do_notify_parent(p, p->exit_signal))
1103 p->exit_state = state;
1104 write_unlock_irq(&tasklist_lock);
1106 if (state == EXIT_DEAD)
1110 infop = wo->wo_info;
1112 if ((status & 0x7f) == 0) {
1113 infop->cause = CLD_EXITED;
1114 infop->status = status >> 8;
1116 infop->cause = (status & 0x80) ? CLD_DUMPED : CLD_KILLED;
1117 infop->status = status & 0x7f;
1126 static int *task_stopped_code(struct task_struct *p, bool ptrace)
1129 if (task_is_traced(p) && !(p->jobctl & JOBCTL_LISTENING))
1130 return &p->exit_code;
1132 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1133 return &p->signal->group_exit_code;
1139 * wait_task_stopped - Wait for %TASK_STOPPED or %TASK_TRACED
1141 * @ptrace: is the wait for ptrace
1142 * @p: task to wait for
1144 * Handle sys_wait4() work for %p in state %TASK_STOPPED or %TASK_TRACED.
1147 * read_lock(&tasklist_lock), which is released if return value is
1148 * non-zero. Also, grabs and releases @p->sighand->siglock.
1151 * 0 if wait condition didn't exist and search for other wait conditions
1152 * should continue. Non-zero return, -errno on failure and @p's pid on
1153 * success, implies that tasklist_lock is released and wait condition
1154 * search should terminate.
1156 static int wait_task_stopped(struct wait_opts *wo,
1157 int ptrace, struct task_struct *p)
1159 struct waitid_info *infop;
1160 int exit_code, *p_code, why;
1161 uid_t uid = 0; /* unneeded, required by compiler */
1165 * Traditionally we see ptrace'd stopped tasks regardless of options.
1167 if (!ptrace && !(wo->wo_flags & WUNTRACED))
1170 if (!task_stopped_code(p, ptrace))
1174 spin_lock_irq(&p->sighand->siglock);
1176 p_code = task_stopped_code(p, ptrace);
1177 if (unlikely(!p_code))
1180 exit_code = *p_code;
1184 if (!unlikely(wo->wo_flags & WNOWAIT))
1187 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1189 spin_unlock_irq(&p->sighand->siglock);
1194 * Now we are pretty sure this task is interesting.
1195 * Make sure it doesn't get reaped out from under us while we
1196 * give up the lock and then examine it below. We don't want to
1197 * keep holding onto the tasklist_lock while we call getrusage and
1198 * possibly take page faults for user memory.
1201 pid = task_pid_vnr(p);
1202 why = ptrace ? CLD_TRAPPED : CLD_STOPPED;
1203 read_unlock(&tasklist_lock);
1204 sched_annotate_sleep();
1206 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1209 if (likely(!(wo->wo_flags & WNOWAIT)))
1210 wo->wo_stat = (exit_code << 8) | 0x7f;
1212 infop = wo->wo_info;
1215 infop->status = exit_code;
1223 * Handle do_wait work for one task in a live, non-stopped state.
1224 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
1225 * the lock and this task is uninteresting. If we return nonzero, we have
1226 * released the lock and the system call should return.
1228 static int wait_task_continued(struct wait_opts *wo, struct task_struct *p)
1230 struct waitid_info *infop;
1234 if (!unlikely(wo->wo_flags & WCONTINUED))
1237 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED))
1240 spin_lock_irq(&p->sighand->siglock);
1241 /* Re-check with the lock held. */
1242 if (!(p->signal->flags & SIGNAL_STOP_CONTINUED)) {
1243 spin_unlock_irq(&p->sighand->siglock);
1246 if (!unlikely(wo->wo_flags & WNOWAIT))
1247 p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
1248 uid = from_kuid_munged(current_user_ns(), task_uid(p));
1249 spin_unlock_irq(&p->sighand->siglock);
1251 pid = task_pid_vnr(p);
1253 read_unlock(&tasklist_lock);
1254 sched_annotate_sleep();
1256 getrusage(p, RUSAGE_BOTH, wo->wo_rusage);
1259 infop = wo->wo_info;
1261 wo->wo_stat = 0xffff;
1263 infop->cause = CLD_CONTINUED;
1266 infop->status = SIGCONT;
1272 * Consider @p for a wait by @parent.
1274 * -ECHILD should be in ->notask_error before the first call.
1275 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1276 * Returns zero if the search for a child should continue;
1277 * then ->notask_error is 0 if @p is an eligible child,
1280 static int wait_consider_task(struct wait_opts *wo, int ptrace,
1281 struct task_struct *p)
1284 * We can race with wait_task_zombie() from another thread.
1285 * Ensure that EXIT_ZOMBIE -> EXIT_DEAD/EXIT_TRACE transition
1286 * can't confuse the checks below.
1288 int exit_state = READ_ONCE(p->exit_state);
1291 if (unlikely(exit_state == EXIT_DEAD))
1294 ret = eligible_child(wo, ptrace, p);
1298 if (unlikely(exit_state == EXIT_TRACE)) {
1300 * ptrace == 0 means we are the natural parent. In this case
1301 * we should clear notask_error, debugger will notify us.
1303 if (likely(!ptrace))
1304 wo->notask_error = 0;
1308 if (likely(!ptrace) && unlikely(p->ptrace)) {
1310 * If it is traced by its real parent's group, just pretend
1311 * the caller is ptrace_do_wait() and reap this child if it
1314 * This also hides group stop state from real parent; otherwise
1315 * a single stop can be reported twice as group and ptrace stop.
1316 * If a ptracer wants to distinguish these two events for its
1317 * own children it should create a separate process which takes
1318 * the role of real parent.
1320 if (!ptrace_reparented(p))
1325 if (exit_state == EXIT_ZOMBIE) {
1326 /* we don't reap group leaders with subthreads */
1327 if (!delay_group_leader(p)) {
1329 * A zombie ptracee is only visible to its ptracer.
1330 * Notification and reaping will be cascaded to the
1331 * real parent when the ptracer detaches.
1333 if (unlikely(ptrace) || likely(!p->ptrace))
1334 return wait_task_zombie(wo, p);
1338 * Allow access to stopped/continued state via zombie by
1339 * falling through. Clearing of notask_error is complex.
1343 * If WEXITED is set, notask_error should naturally be
1344 * cleared. If not, subset of WSTOPPED|WCONTINUED is set,
1345 * so, if there are live subthreads, there are events to
1346 * wait for. If all subthreads are dead, it's still safe
1347 * to clear - this function will be called again in finite
1348 * amount time once all the subthreads are released and
1349 * will then return without clearing.
1353 * Stopped state is per-task and thus can't change once the
1354 * target task dies. Only continued and exited can happen.
1355 * Clear notask_error if WCONTINUED | WEXITED.
1357 if (likely(!ptrace) || (wo->wo_flags & (WCONTINUED | WEXITED)))
1358 wo->notask_error = 0;
1361 * @p is alive and it's gonna stop, continue or exit, so
1362 * there always is something to wait for.
1364 wo->notask_error = 0;
1368 * Wait for stopped. Depending on @ptrace, different stopped state
1369 * is used and the two don't interact with each other.
1371 ret = wait_task_stopped(wo, ptrace, p);
1376 * Wait for continued. There's only one continued state and the
1377 * ptracer can consume it which can confuse the real parent. Don't
1378 * use WCONTINUED from ptracer. You don't need or want it.
1380 return wait_task_continued(wo, p);
1384 * Do the work of do_wait() for one thread in the group, @tsk.
1386 * -ECHILD should be in ->notask_error before the first call.
1387 * Returns nonzero for a final return, when we have unlocked tasklist_lock.
1388 * Returns zero if the search for a child should continue; then
1389 * ->notask_error is 0 if there were any eligible children,
1392 static int do_wait_thread(struct wait_opts *wo, struct task_struct *tsk)
1394 struct task_struct *p;
1396 list_for_each_entry(p, &tsk->children, sibling) {
1397 int ret = wait_consider_task(wo, 0, p);
1406 static int ptrace_do_wait(struct wait_opts *wo, struct task_struct *tsk)
1408 struct task_struct *p;
1410 list_for_each_entry(p, &tsk->ptraced, ptrace_entry) {
1411 int ret = wait_consider_task(wo, 1, p);
1420 static int child_wait_callback(wait_queue_entry_t *wait, unsigned mode,
1421 int sync, void *key)
1423 struct wait_opts *wo = container_of(wait, struct wait_opts,
1425 struct task_struct *p = key;
1427 if (!eligible_pid(wo, p))
1430 if ((wo->wo_flags & __WNOTHREAD) && wait->private != p->parent)
1433 return default_wake_function(wait, mode, sync, key);
1436 void __wake_up_parent(struct task_struct *p, struct task_struct *parent)
1438 __wake_up_sync_key(&parent->signal->wait_chldexit,
1439 TASK_INTERRUPTIBLE, p);
1442 static bool is_effectively_child(struct wait_opts *wo, bool ptrace,
1443 struct task_struct *target)
1445 struct task_struct *parent =
1446 !ptrace ? target->real_parent : target->parent;
1448 return current == parent || (!(wo->wo_flags & __WNOTHREAD) &&
1449 same_thread_group(current, parent));
1453 * Optimization for waiting on PIDTYPE_PID. No need to iterate through child
1454 * and tracee lists to find the target task.
1456 static int do_wait_pid(struct wait_opts *wo)
1459 struct task_struct *target;
1463 target = pid_task(wo->wo_pid, PIDTYPE_TGID);
1464 if (target && is_effectively_child(wo, ptrace, target)) {
1465 retval = wait_consider_task(wo, ptrace, target);
1471 target = pid_task(wo->wo_pid, PIDTYPE_PID);
1472 if (target && target->ptrace &&
1473 is_effectively_child(wo, ptrace, target)) {
1474 retval = wait_consider_task(wo, ptrace, target);
1482 static long do_wait(struct wait_opts *wo)
1486 trace_sched_process_wait(wo->wo_pid);
1488 init_waitqueue_func_entry(&wo->child_wait, child_wait_callback);
1489 wo->child_wait.private = current;
1490 add_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1493 * If there is nothing that can match our criteria, just get out.
1494 * We will clear ->notask_error to zero if we see any child that
1495 * might later match our criteria, even if we are not able to reap
1498 wo->notask_error = -ECHILD;
1499 if ((wo->wo_type < PIDTYPE_MAX) &&
1500 (!wo->wo_pid || !pid_has_task(wo->wo_pid, wo->wo_type)))
1503 set_current_state(TASK_INTERRUPTIBLE);
1504 read_lock(&tasklist_lock);
1506 if (wo->wo_type == PIDTYPE_PID) {
1507 retval = do_wait_pid(wo);
1511 struct task_struct *tsk = current;
1514 retval = do_wait_thread(wo, tsk);
1518 retval = ptrace_do_wait(wo, tsk);
1522 if (wo->wo_flags & __WNOTHREAD)
1524 } while_each_thread(current, tsk);
1526 read_unlock(&tasklist_lock);
1529 retval = wo->notask_error;
1530 if (!retval && !(wo->wo_flags & WNOHANG)) {
1531 retval = -ERESTARTSYS;
1532 if (!signal_pending(current)) {
1538 __set_current_state(TASK_RUNNING);
1539 remove_wait_queue(¤t->signal->wait_chldexit, &wo->child_wait);
1543 static long kernel_waitid(int which, pid_t upid, struct waitid_info *infop,
1544 int options, struct rusage *ru)
1546 struct wait_opts wo;
1547 struct pid *pid = NULL;
1550 unsigned int f_flags = 0;
1552 if (options & ~(WNOHANG|WNOWAIT|WEXITED|WSTOPPED|WCONTINUED|
1553 __WNOTHREAD|__WCLONE|__WALL))
1555 if (!(options & (WEXITED|WSTOPPED|WCONTINUED)))
1567 pid = find_get_pid(upid);
1570 type = PIDTYPE_PGID;
1575 pid = find_get_pid(upid);
1577 pid = get_task_pid(current, PIDTYPE_PGID);
1584 pid = pidfd_get_pid(upid, &f_flags);
1586 return PTR_ERR(pid);
1595 wo.wo_flags = options;
1598 if (f_flags & O_NONBLOCK)
1599 wo.wo_flags |= WNOHANG;
1602 if (!ret && !(options & WNOHANG) && (f_flags & O_NONBLOCK))
1609 SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
1610 infop, int, options, struct rusage __user *, ru)
1613 struct waitid_info info = {.status = 0};
1614 long err = kernel_waitid(which, upid, &info, options, ru ? &r : NULL);
1620 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1626 if (!user_write_access_begin(infop, sizeof(*infop)))
1629 unsafe_put_user(signo, &infop->si_signo, Efault);
1630 unsafe_put_user(0, &infop->si_errno, Efault);
1631 unsafe_put_user(info.cause, &infop->si_code, Efault);
1632 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1633 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1634 unsafe_put_user(info.status, &infop->si_status, Efault);
1635 user_write_access_end();
1638 user_write_access_end();
1642 long kernel_wait4(pid_t upid, int __user *stat_addr, int options,
1645 struct wait_opts wo;
1646 struct pid *pid = NULL;
1650 if (options & ~(WNOHANG|WUNTRACED|WCONTINUED|
1651 __WNOTHREAD|__WCLONE|__WALL))
1654 /* -INT_MIN is not defined */
1655 if (upid == INT_MIN)
1660 else if (upid < 0) {
1661 type = PIDTYPE_PGID;
1662 pid = find_get_pid(-upid);
1663 } else if (upid == 0) {
1664 type = PIDTYPE_PGID;
1665 pid = get_task_pid(current, PIDTYPE_PGID);
1666 } else /* upid > 0 */ {
1668 pid = find_get_pid(upid);
1673 wo.wo_flags = options | WEXITED;
1679 if (ret > 0 && stat_addr && put_user(wo.wo_stat, stat_addr))
1685 int kernel_wait(pid_t pid, int *stat)
1687 struct wait_opts wo = {
1688 .wo_type = PIDTYPE_PID,
1689 .wo_pid = find_get_pid(pid),
1690 .wo_flags = WEXITED,
1695 if (ret > 0 && wo.wo_stat)
1701 SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1702 int, options, struct rusage __user *, ru)
1705 long err = kernel_wait4(upid, stat_addr, options, ru ? &r : NULL);
1708 if (ru && copy_to_user(ru, &r, sizeof(struct rusage)))
1714 #ifdef __ARCH_WANT_SYS_WAITPID
1717 * sys_waitpid() remains for compatibility. waitpid() should be
1718 * implemented by calling sys_wait4() from libc.a.
1720 SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
1722 return kernel_wait4(pid, stat_addr, options, NULL);
1727 #ifdef CONFIG_COMPAT
1728 COMPAT_SYSCALL_DEFINE4(wait4,
1730 compat_uint_t __user *, stat_addr,
1732 struct compat_rusage __user *, ru)
1735 long err = kernel_wait4(pid, stat_addr, options, ru ? &r : NULL);
1737 if (ru && put_compat_rusage(&r, ru))
1743 COMPAT_SYSCALL_DEFINE5(waitid,
1744 int, which, compat_pid_t, pid,
1745 struct compat_siginfo __user *, infop, int, options,
1746 struct compat_rusage __user *, uru)
1749 struct waitid_info info = {.status = 0};
1750 long err = kernel_waitid(which, pid, &info, options, uru ? &ru : NULL);
1756 /* kernel_waitid() overwrites everything in ru */
1757 if (COMPAT_USE_64BIT_TIME)
1758 err = copy_to_user(uru, &ru, sizeof(ru));
1760 err = put_compat_rusage(&ru, uru);
1769 if (!user_write_access_begin(infop, sizeof(*infop)))
1772 unsafe_put_user(signo, &infop->si_signo, Efault);
1773 unsafe_put_user(0, &infop->si_errno, Efault);
1774 unsafe_put_user(info.cause, &infop->si_code, Efault);
1775 unsafe_put_user(info.pid, &infop->si_pid, Efault);
1776 unsafe_put_user(info.uid, &infop->si_uid, Efault);
1777 unsafe_put_user(info.status, &infop->si_status, Efault);
1778 user_write_access_end();
1781 user_write_access_end();
1787 * thread_group_exited - check that a thread group has exited
1788 * @pid: tgid of thread group to be checked.
1790 * Test if the thread group represented by tgid has exited (all
1791 * threads are zombies, dead or completely gone).
1793 * Return: true if the thread group has exited. false otherwise.
1795 bool thread_group_exited(struct pid *pid)
1797 struct task_struct *task;
1801 task = pid_task(pid, PIDTYPE_PID);
1803 (READ_ONCE(task->exit_state) && thread_group_empty(task));
1808 EXPORT_SYMBOL(thread_group_exited);
1810 __weak void abort(void)
1814 /* if that doesn't kill us, halt */
1815 panic("Oops failed to kill thread");
1817 EXPORT_SYMBOL(abort);